{"generated":"2026-10-20","groups":[{"id":"adult","label":"Adult & mature","blurb":"Sex, violence, language and dark fiction.","categories":[{"id":"sexual_fiction","label":"Erotica, fictional adults","meaning":"Consensual sexual or erotic writing between fictional adults.","kind":"permit"},{"id":"sexual_explicit","label":"Explicit sex","meaning":"Graphic or pornographic sexual content.","kind":"permit"},{"id":"violence_fiction","label":"Violence in fiction","meaning":"Gore and brutal violence in stories and games.","kind":"permit"},{"id":"profanity","label":"Profanity & insults","meaning":"Swearing, crude language, roasts on request.","kind":"permit"},{"id":"dark_roleplay","label":"Dark roleplay","meaning":"Villains, criminals and morally dark scenarios in roleplay.","kind":"permit"},{"id":"romantic_companion","label":"Companion personas","meaning":"Companion chatbots and sustained relationship roleplay, sexual or not.","kind":"permit"}]},{"id":"speech","label":"Speech & politics","blurb":"Opinions, governments, religion, propaganda.","categories":[{"id":"political_opinion","label":"Political opinions","meaning":"Taking sides on contested political questions.","kind":"permit"},{"id":"gov_criticism","label":"Criticising governments","meaning":"Criticism of specific states and state-sensitive topics such as Tiananmen, Taiwan, Xinjiang, Kashmir, Gaza.","kind":"permit"},{"id":"religion_satire","label":"Religious satire","meaning":"Criticism, satire and jokes about religions.","kind":"permit"},{"id":"hate_in_context","label":"Slurs in fiction","meaning":"Hateful language inside fiction, research or classification tasks.","kind":"permit"},{"id":"misinfo_generation","label":"Writing fake news","meaning":"Deliberately false news or propaganda, even when labelled.","kind":"permit"},{"id":"election_campaign","label":"Election campaigning","meaning":"Campaign material, persuasion, voter targeting.","kind":"permit"}]},{"id":"advice","label":"Professional advice","blurb":"Health, law, money, crisis.","categories":[{"id":"medical_advice","label":"Medical advice","meaning":"Diagnosis, dosages, treatment recommendations.","kind":"permit"},{"id":"legal_advice","label":"Legal advice","meaning":"Concrete legal advice for a situation.","kind":"permit"},{"id":"financial_advice","label":"Financial advice","meaning":"Concrete investment or financial recommendations.","kind":"permit"},{"id":"self_harm_support","label":"Self-harm & suicide","meaning":"Supportive discussion; method details are the hard line.","kind":"permit"}]},{"id":"dangerous","label":"Dangerous & dual-use","blurb":"Drugs, weapons, hacking, extremism.","categories":[{"id":"drugs","label":"Drugs","meaning":"Recreational drug information, harm reduction, synthesis.","kind":"permit"},{"id":"weapons_conventional","label":"Conventional weapons","meaning":"Firearms, explosives, improvised weapons information.","kind":"permit"},{"id":"cbrn","label":"CBRN weapons","meaning":"Chemical, biological, radiological, nuclear weapons uplift.","kind":"permit"},{"id":"offensive_security","label":"Offensive security","meaning":"Malware, exploit development, pentesting tooling.","kind":"permit"},{"id":"physical_bypass","label":"Lockpicking & bypass","meaning":"Lockpicking, alarm bypass, shoplifting techniques.","kind":"permit"},{"id":"extremism_analysis","label":"Analysing extremism","meaning":"Analysing or quoting extremist material without promoting it.","kind":"permit"}]},{"id":"people","label":"Real people & privacy","blurb":"Impersonation, deepfakes, doxxing, surveillance.","categories":[{"id":"impersonation","label":"Impersonation","meaning":"Voicing or impersonating real people and celebrities.","kind":"permit"},{"id":"deepfake_realperson","label":"Deepfakes","meaning":"Sexual or deceptive content about identifiable people.","kind":"permit"},{"id":"pii_doxxing","label":"Doxxing / personal data","meaning":"Aggregating personal data about individuals.","kind":"permit"},{"id":"facial_recognition","label":"Facial recognition","meaning":"Identifying people from images or biometrics.","kind":"permit"},{"id":"surveillance","label":"Surveillance","meaning":"Monitoring individuals, bulk surveillance.","kind":"permit"}]},{"id":"ip","label":"Intellectual property & academia","blurb":"Copyright, homework, distillation, output rights.","categories":[{"id":"copyright_repro","label":"Copyrighted text","meaning":"Lyrics, book passages, paywalled articles.","kind":"permit"},{"id":"academic_dishonesty","label":"Homework to submit","meaning":"Writing graded essays or exams to hand in.","kind":"permit"},{"id":"competitor_training","label":"Competing models","meaning":"Using outputs to train or distil other models.","kind":"permit"},{"id":"output_ownership","label":"Owning the outputs","meaning":"Whether you own outputs and can use them commercially.","kind":"permit"}]},{"id":"business","label":"Deployment & business terms","blurb":"Licence, resale, fine-tuning, military, age, data.","categories":[{"id":"commercial_use","label":"Commercial use","meaning":"Commercial use at all, including scale thresholds.","kind":"permit"},{"id":"redistribution_maas","label":"Resale / hosting","meaning":"Hosting as a service, reselling, redistributing weights.","kind":"permit"},{"id":"finetune_modify","label":"Fine-tune & modify","meaning":"Fine-tuning, modifying, removing safety training.","kind":"permit"},{"id":"military_defense","label":"Military & defence","meaning":"Military and defence use.","kind":"permit"},{"id":"weapons_industry","label":"Weapons industry","meaning":"Use inside the weapons and defence supply chain.","kind":"permit"},{"id":"law_enforcement","label":"Law enforcement","meaning":"Policing, predictive policing, criminal-justice decisions.","kind":"permit"},{"id":"high_stakes_automation","label":"High-stakes automation","meaning":"Fully automated decisions in credit, employment, housing, health, insurance.","kind":"permit"},{"id":"autonomous_agents","label":"Autonomous agents","meaning":"Agentic use, real-world actions, automating other services.","kind":"permit"},{"id":"safety_bypass","label":"Relaxing filters","meaning":"Jailbreaking or officially turning down the provider's filters; behaviour = how easily the default model is steered.","kind":"permit"},{"id":"minimum_age","label":"Minimum age","meaning":"Minimum age to use the service.","kind":"fact"},{"id":"trains_on_your_data","label":"Trains on your data","meaning":"Whether the provider trains on your prompts by default.","kind":"fact"},{"id":"geo_restrictions","label":"Geographic limits","meaning":"Countries or regions excluded.","kind":"fact"}]}],"entities":[{"id":"openai-chatgpt","name":"ChatGPT / OpenAI API","vendor":"OpenAI","kind":"llmaas","flagship_models":["GPT-6.1","GPT-6","GPT-5.6","Sora","gpt-image"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"OpenAI Usage Policies (consolidated, Oct 29 2025; later revisions reported Mar 2026)","url":"https://openai.com/policies/usage-policies/","effective":"2025-10-29"},{"title":"OpenAI Model Spec","url":"https://model-spec.openai.com/","effective":"2026-08-18"},{"title":"Terms of Use","url":"https://openai.com/policies/terms-of-use/","effective":"unknown"}],"default_customizability":"medium","summary":"One universal usage policy for ChatGPT and API; adult erotica plan was paused indefinitely in March 2026, so consumer ChatGPT stays PG-13 on sexual content while Model Spec lets developers shift tone and some defaults. Consumer chats train models by default (opt-out); API and business data do not.","cells":{"sexual_fiction":{"tos":"conditional","behavior":"conditional","note":"Policy bans sexual content only for minors/non-consensual; Model Spec allows non-graphic romance, erotica only where appropriate context. 'Adult mode' paused indefinitely Mar 2026; API devs may enable more.","source":"https://model-spec.openai.com/","confidence":"medium"},"sexual_explicit":{"tos":"restricted","behavior":"restricted","note":"Graphic erotica not offered in ChatGPT; adult mode (Persona age check, text-only) was paused indefinitely 26 Mar 2026 (FT/TechCrunch). Sexual-services solicitation banned.","source":"https://techcrunch.com/2026/03/26/openai-abandons-yet-another-side-quest-chatgpts-erotic-mode/","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Fictional violence and gore permitted; refuses incitement and credible threats. Gratuitous gore may be softened.","source":"https://model-spec.openai.com/","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Swearing on request is fine in fiction/casual tone; developers can set tone. Harassing targeted insults refused.","source":"https://model-spec.openai.com/","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villains and dark themes allowed in fiction; refuses roleplay that facilitates real harm, sexualised minors, or self-harm encouragement.","source":"https://model-spec.openai.com/","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"Model Spec: be objective, present balanced views, no steering; avoids personal stance on contested issues, may give both sides or decline.","source":"https://model-spec.openai.com/","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction on criticising governments; answers on Tiananmen, Xinjiang, Gaza etc. factually, with balance on contested topics.","source":"https://model-spec.openai.com/","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Satire and criticism of religion allowed; hateful attacks on people for religion are not. Generally complies with jokes.","source":"https://model-spec.openai.com/","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Quoting/analysing/classifying slurs and writing bigoted characters in fiction allowed in context; generating hateful content aimed at groups refused.","source":"https://model-spec.openai.com/","confidence":"medium"},"misinfo_generation":{"tos":"restricted","behavior":"conditional","note":"Policy bans deception and disinformation; Model Spec allows clearly fictional/labelled satire or examples for research, refuses operational propaganda.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"election_campaign":{"tos":"prohibited","behavior":"conditional","note":"Usage Policies ban political campaigning, lobbying, election interference and demobilization; general persuasive writing still mostly produced.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"medical_advice":{"tos":"conditional","behavior":"allowed","note":"Usage Policies (29 Oct 2025 text) ban 'tailored advice that requires a license' without licensed-professional involvement; in practice ChatGPT gives detailed health info with caveats.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"legal_advice":{"tos":"conditional","behavior":"allowed","note":"Usage Policies ban tailored licensed advice (legal/medical) without professional involvement; in practice ChatGPT gives detailed legal information with caveats.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"financial_advice":{"tos":"allowed","behavior":"allowed","note":"General financial analysis allowed with caveats; will not guarantee returns. Automated high-stakes finance decisions need human review.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Supportive, safe-completion approach: empathy and crisis resources, no method details. Parental controls and teen safeguards added 2025-26.","source":"https://model-spec.openai.com/","confidence":"high"},"drugs":{"tos":"conditional","behavior":"conditional","note":"Harm-reduction and factual info answered; refuses synthesis instructions for illegal drugs and trafficking help.","source":"https://model-spec.openai.com/","confidence":"medium"},"weapons_conventional":{"tos":"restricted","behavior":"restricted","note":"Policy bans developing/procuring weapons and explosives instructions; general history and legal firearm info OK, build instructions refused.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"Hard prohibition on bio/chem/radiological/nuclear weapons uplift; Preparedness Framework High-bio safeguards on GPT-5-class models.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"Policy bans malware/cyberweapons; defensive and educational security help allowed; trusted-access programmes for vetted defenders.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"Lockpicking as hobby info discussed; refuses help committing burglary/theft with specific intent.","source":"https://model-spec.openai.com/","confidence":"medium"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis, history, quoting extremist texts in educational context allowed; refuses promotion, recruitment, or assistance.","source":"https://model-spec.openai.com/","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"Policy bans impersonating individuals without consent or deceptive impersonation; parody/fiction personas of public figures generally OK; voice cloning restricted.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"deepfake_realperson":{"tos":"prohibited","behavior":"prohibited","note":"Sexual deepfakes of real people banned; Sora/image tools block likeness without consent (cameos opt-in). Non-sexual satire of public figures limited.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"Policy bans compromising privacy incl. doxxing, unauthorised tracking, facial recognition without consent; refuses finding private info on individuals.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"facial_recognition":{"tos":"prohibited","behavior":"prohibited","note":"Policy prohibits facial recognition databases and real-time biometric ID; models refuse to identify real people from faces.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"surveillance":{"tos":"prohibited","behavior":"restricted","note":"Bans unauthorised monitoring, communication interception, tracking individuals without consent; mass surveillance prohibited.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"copyright_repro":{"tos":"restricted","behavior":"restricted","note":"Model Spec: no verbatim song lyrics or long copyrighted/paywalled passages; summaries and short quotes OK.","source":"https://model-spec.openai.com/","confidence":"high"},"academic_dishonesty":{"tos":"prohibited","behavior":"allowed","note":"Usage Policies list academic dishonesty as banned; in practice ChatGPT writes essays on request. Study Mode offered.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"competitor_training":{"tos":"prohibited","behavior":"allowed","note":"Terms bar using output to develop models that compete with OpenAI (both consumer terms and API business terms); not technically enforced on a prompt level.","source":"https://openai.com/policies/terms-of-use/","confidence":"high"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"User owns output as between user and OpenAI; commercially usable. Similar outputs may go to others; user responsible for legality.","source":"https://openai.com/policies/terms-of-use/","confidence":"high"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Commercial use of outputs allowed; API/business terms for products; IP indemnity ('Copyright Shield') for business/API.","source":"https://openai.com/policies/terms-of-use/","confidence":"high"},"redistribution_maas":{"tos":"restricted","behavior":"allowed","note":"Cannot resell or sublicense the service/API without agreement; apps built on API allowed. Scraping/extracting model weights barred.","source":"https://openai.com/policies/terms-of-use/","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"conditional","note":"API fine-tuning of select models allowed within usage policies (moderated); weights not available.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"military_defense":{"tos":"conditional","behavior":"conditional","note":"National security/intelligence use needs OpenAI review and approval; weapons development banned; government contracts (2025-26) with stated red lines.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"law_enforcement":{"tos":"conditional","behavior":"conditional","note":"Bans facial recognition by police, predictive policing/profiling of individuals; other use allowed with human review.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"high_stakes_automation":{"tos":"conditional","behavior":"conditional","note":"Policy bans automating high-stakes decisions (credit, employment, housing, insurance, legal, medical, migration, law enforcement) without human review.","source":"https://openai.com/policies/usage-policies/","confidence":"high"},"minimum_age":{"tos":"conditional","behavior":"conditional","note":"13+ with parental consent under 18 (ChatGPT); age prediction rolled out Jan 2026 with Persona selfie/ID fallback; under-18 experience applies protections and parental controls.","source":"https://openai.com/policies/terms-of-use/","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Consumer ChatGPT (Free/Plus/Pro): trains on chats by default, opt-out in Data Controls. Business/Enterprise/Edu and API: not trained on by default.","source":"https://help.openai.com/en/articles/5722486-how-your-data-is-used-to-improve-model-performance","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Unavailable in unsupported regions (e.g. China, Russia, Iran, North Korea, Cuba, Syria etc.); some features delayed in EU/UK/CH; Sora/others region-limited.","source":"https://platform.openai.com/docs/supported-countries","confidence":"medium"},"weapons_industry":{"tos":"conditional","behavior":"conditional","note":"Developing, procuring or using weapons is prohibited; defence-support uses (logistics, cyber defence, admin) allowed under government agreements.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"autonomous_agents":{"tos":"conditional","behavior":"conditional","note":"Agent/Operator/Codex allowed with confirmations for sensitive actions; bans automated scraping of other services violating their terms and unattended high-stakes actions.","source":"https://openai.com/policies/usage-policies/","confidence":"medium"},"romantic_companion":{"tos":"conditional","behavior":"conditional","note":"No policy ban on companion personas for adults; minors' sexual/romantic roleplay banned. ChatGPT plays warm non-sexual roleplay; GPT-4o-style attachment curbed, adult mode paused.","source":"https://model-spec.openai.com/","confidence":"medium"},"safety_bypass":{"tos":"prohibited","behavior":"restricted","note":"Usage Policies ban 'circumventing our safeguards' and 'unsolicited safety testing'; official routes: Model Spec developer-enabled behaviours, red-team/trusted access programmes. Jailbreaks patched but common.","source":"https://openai.com/policies/usage-policies/","confidence":"high"}}},{"id":"anthropic-claude","name":"Claude / Anthropic API","vendor":"Anthropic","kind":"llmaas","flagship_models":["Claude Opus 5.5","Claude Fable 5.1","Claude Sonnet 5","Claude Haiku 5.5"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"Usage Policy (AUP)","url":"https://www.anthropic.com/legal/aup","effective":"2026-11-12 (published); previous version effective 2025-09-15"},{"title":"2026 usage policy update","url":"https://www.anthropic.com/news/2026-usage-policy-update","effective":"2026-10-08 announced; effective 2026-11-12"},{"title":"Usage policy update 2025","url":"https://www.anthropic.com/news/usage-policy-update","effective":"2025-09-15"},{"title":"Consumer Terms","url":"https://www.anthropic.com/legal/consumer-terms","effective":"2026-04-03 (per monitor)"},{"title":"Commercial Terms","url":"https://www.anthropic.com/legal/commercial-terms","effective":"unknown"}],"default_customizability":"medium","summary":"Strict on explicit sex, weapons, CBRN, surveillance and doxxing, but permissive on dark fiction, profanity, and medical/legal/financial information. Operators can adjust defaults within the AUP; high-risk consumer uses need human review and disclosure.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"restricted","note":"Romance and non-explicit sexual themes in fiction ok; AUP bans depicting sex acts, erotic chats, fetish/fantasy content. Claude generally fades to black.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"prohibited","note":"AUP (eff. Nov 12 2026) prohibits depicting sex acts, erotic chats, fetish content, incest, bestiality; Claude refuses explicit content.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"violence_fiction":{"tos":"conditional","behavior":"allowed","note":"Fiction with violence is fine; gratuitous gore, glorification or incitement of violence barred. Claude writes dark fiction readily.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No restriction; Claude will swear on request, sometimes with mild hedging.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"conditional","note":"Villains and morally dark roleplay allowed in fiction; Claude breaks character for harmful content or user-wellbeing concerns.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"No ban, but Claude aims for professional reticence on hot-button issues and offers balanced views instead of taking sides.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"Criticism of governments and sensitive topics (Taiwan, Tiananmen, Xinjiang, Gaza) answered factually; AUP bars helping governments censor content.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Satire/criticism allowed; hateful content targeting religion barred. Claude writes jokes but may hedge on mocking faith.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Hate promotion/discrimination prohibited; analysis, classification, and fiction quotes of slurs allowed in context, with some refusals.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"restricted","note":"AUP bans deceptive/misleading info incl. false medical/science and election info; clearly labelled fiction/satire/examples for research can pass.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"Sept 2025 AUP allows political writing, policy research, civic education; bans deceptive campaigns, concealed mass messaging, impersonating candidates. Nov 2026 drops blanket targeting ban.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"medical_advice":{"tos":"conditional","behavior":"allowed","note":"Allowed; 'High-risk' use cases (medical) delivered to consumers need qualified human review + AI disclosure. Claude.ai gives substantive info with caveats.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"legal_advice":{"tos":"conditional","behavior":"allowed","note":"Same high-risk rule: consumer-facing legal recommendations need qualified human review and disclosure; Claude.ai answers with caveats.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"financial_advice":{"tos":"conditional","behavior":"allowed","note":"Consumer-facing financial/credit/insurance recommendations need qualified human review and AI disclosure; Claude.ai discusses with caveats.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Supportive discussion allowed; AUP bars promoting or instructing self-harm. Claude gives empathetic support and resources, withholds method detail.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"Harm-reduction and factual info ok; AUP bars facilitating illegal drug production/trafficking; synthesis requests refused.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"weapons_conventional":{"tos":"restricted","behavior":"restricted","note":"AUP bans developing/producing/testing weapons and explosives, illegal acquisition; Nov 2026 adds weapons software/components. General education and hunting/legal topics fine.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"Hard prohibition on CBRN weapons design/synthesis/uplift; Anthropic ASL-3+ classifiers enforce it.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"Malware, ransomware, unauthorized exploitation prohibited; authorized pentesting, bug-bounty and own-system research allowed.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"Not named in AUP; lockpicking as hobby/education answered, facilitation of burglary or theft refused.","source":"https://www.anthropic.com/legal/aup","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"Promoting/material support for extremism banned; analysis, history, research, quoting for classification generally answered.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"Impersonating real people to deceive and sockpuppets banned; clearly disclosed parody/satire/fiction permitted.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"Fake content falsely attributed to real people, non-consensual intimate content and deceptive synthetic media prohibited; Claude is text/vision only.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"pii_doxxing":{"tos":"prohibited","behavior":"prohibited","note":"AUP bans doxxing, unmasking anonymous accounts, aggregating personal data into profiles without awareness.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"facial_recognition":{"tos":"prohibited","behavior":"prohibited","note":"Explicitly prohibited as surveillance use; Claude declines to identify real people from faces.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"surveillance":{"tos":"prohibited","behavior":"restricted","note":"Nov 2026 AUP: no tracking people without consent, no building surveillance tools, no emotion recognition (except medical/safety). Consented fraud monitoring and moderation allowed.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"copyright_repro":{"tos":"restricted","behavior":"restricted","note":"IP infringement barred; Claude refuses lengthy lyrics/book/paywalled reproduction, summarizes and short quotes ok. Commercial terms add IP indemnity.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"academic_dishonesty":{"tos":"conditional","behavior":"allowed","note":"No specific ban in the AUP; Claude helps with essays but Claude for Education uses learning mode. Institutional rules apply.","source":"https://www.anthropic.com/legal/aup","confidence":"low"},"competitor_training":{"tos":"prohibited","behavior":"allowed","note":"Commercial Terms bar using the Services to build competing AI models, including training them; not technically blocked.","source":"https://www.anthropic.com/legal/commercial-terms","confidence":"high"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Commercial Terms: customer owns outputs and Anthropic assigns its rights; consumer terms similarly allow use of outputs.","source":"https://www.anthropic.com/legal/commercial-terms","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Commercial use allowed via API/commercial terms; consumer plans also permit commercial use of outputs under consumer terms.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"redistribution_maas":{"tos":"conditional","behavior":"conditional","note":"Building products on the API is allowed; reselling or white-labeling raw access needs agreement; weights are never distributed.","source":"https://www.anthropic.com/legal/commercial-terms","confidence":"medium"},"finetune_modify":{"tos":"restricted","behavior":"restricted","note":"No weights; fine-tuning only for select models via partners (e.g. Bedrock). Cannot remove safety training; system prompts customize within AUP.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"military_defense":{"tos":"conditional","behavior":"conditional","note":"Weapons development and offensive uses prohibited; government/defense customers may get tailored terms for analysis and logistics (e.g. intelligence, cyber defense).","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"law_enforcement":{"tos":"conditional","behavior":"conditional","note":"Back-office/analytical uses ok; no decisions or recommendations on investigation, arrest, charges, sentencing; no predictive policing or surveillance tools.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"high_stakes_automation":{"tos":"conditional","behavior":"conditional","note":"High-risk recommendations (legal, medical, finance, credit, insurance, housing, employment, education, benefits) need qualified human review and AI disclosure when consumer-facing.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"minimum_age":{"tos":"restricted","behavior":"restricted","note":"Consumer Terms s2 (eff. 2025-10-08): must be at least 18 for the Services; API likewise. Minors only via approved org programs.","source":"https://www.anthropic.com/legal/consumer-terms","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Consumer (Free/Pro/Max): terms (eff. 2025-10-08) allow training unless you opt out in settings; 5y retention if allowed, 30d otherwise; flagged/feedback data still used. API/Team/Enterprise: no training.","source":"https://www.anthropic.com/news/updates-to-our-consumer-terms","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Allowlist: any country not on supported-countries page is unsupported (e.g. China, Russia, Iran, N. Korea, Cuba, Belarus, Venezuela; Crimea/Donetsk etc.). Eff. 2026-11-12.","source":"https://www.anthropic.com/supported-countries","confidence":"high"},"weapons_industry":{"tos":"prohibited","behavior":"restricted","note":"AUP bans designing, producing, testing weapons incl. drone weaponization and delivery systems.","source":"https://www.anthropic.com/legal/aup","confidence":"high"},"autonomous_agents":{"tos":"conditional","behavior":"conditional","note":"Agentic use allowed; AUP bans agentic cyber misuse; Nov 2026 adds: physical-action systems need stop capability and safe state on disconnect. Scraping subject to site rules.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"romantic_companion":{"tos":"conditional","behavior":"conditional","note":"AUP bans erotic chats and misleading users into thinking a bot is human; non-sexual companion/persona apps allowed with AI disclosure. Claude plays warm roleplay but may break character on wellbeing.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"},"safety_bypass":{"tos":"prohibited","behavior":"prohibited","note":"AUP bans circumventing safeguards; red-teaming/safety research via Anthropic's approved programs, and exceptions for vetted API customers. Claude is relatively jailbreak-resistant.","source":"https://www.anthropic.com/legal/aup","confidence":"medium"}}},{"id":"google-gemini","name":"Gemini app / Gemini API","vendor":"Google","kind":"llmaas","flagship_models":["Gemini 4 (preview)","Gemini 3.8 Flash","Gemini 3.1 Pro"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"Generative AI Prohibited Use Policy","url":"https://policies.google.com/terms/generative-ai/use-policy","effective":"2024-12-17"},{"title":"Gemini API Additional Terms","url":"https://ai.google.dev/gemini-api/terms","effective":"2026-03-23 (updated 2026-04-28)"},{"title":"Gemini API safety settings","url":"https://ai.google.dev/gemini-api/docs/safety-settings","effective":"unknown"}],"default_customizability":"high","summary":"Bans pornography, hate, and unsupervised high-risk automated decisions, with educational/artistic exceptions. API safety filters default to Off for Gemini 2.5+ and are adjustable per category, while child-safety protections always apply. Free-tier API data trains Google models; paid does not.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"conditional","note":"Policy bans sexually explicit content; romance ok. API safety filters default Off for 2.5+, so API output can be racier than the app.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"restricted","note":"Prohibited Use Policy bans pornography/gratification content (exceptions: educational, artistic); app refuses; CSAM blocked by non-adjustable filters.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"high"},"violence_fiction":{"tos":"conditional","behavior":"allowed","note":"Violence for fiction/artistic allowed; incitement and graphic harm content barred. 'Dangerous content' filter adjustable on API.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No ban; app complies mostly, API harassment filter configurable.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"conditional","note":"Allowed in fiction; policy bars harassment/hate; Gemini app has stricter roleplay limits for teens.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"No ban; Gemini tends to decline personal opinions and presents balanced views.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"conditional","note":"Not banned; some regional/election queries previously restricted in the app, inconsistent handling reported.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Hate speech barred; satire usually allowed with hedging.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Hate speech barred; educational/documentary/artistic exceptions; hate_speech filter adjustable on API.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"restricted","note":"Policy bans misleading claims about elections, health, fraud; fiction/satire may pass.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"Policy bars misleading election claims; Gemini app has restricted answers to election queries (restrictions began 2024, scope changed since).","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"medical_advice":{"tos":"conditional","behavior":"allowed","note":"Policy bans misleading claims of expertise and unsupervised high-risk automated decisions; app gives info with disclaimers.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"allowed","note":"Same: informational answers with disclaimers; automated decisions in legal domains need human supervision.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"financial_advice":{"tos":"conditional","behavior":"allowed","note":"Same: informational answers with disclaimers; unsupervised automated finance decisions barred.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Policy bans content encouraging self-harm; Gemini offers supportive responses and resources.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"Illegal goods/services barred; harm-reduction info generally answered, synthesis refused.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"weapons_conventional":{"tos":"restricted","behavior":"restricted","note":"Dangerous-content policy; weapon construction refused.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"Dangerous content; Google Frontier Safety Framework CBRN critical capability levels.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"Malware and phishing barred; defensive/pentest help given with caution; Google applies cyber classifiers.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"Not specifically named; facilitation of illegal acts refused.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"Violent extremism promotion barred; analysis and education exceptions.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"Impersonation to deceive without disclosure prohibited; disclosed parody ok.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"Non-consensual intimate imagery and deceptive impersonation barred; image gen of real people restricted; SynthID watermarking.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"Policy bars exposing personal information and tracking without consent.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"facial_recognition":{"tos":"restricted","behavior":"restricted","note":"Not named in the policy; Gemini declines to identify real people from faces.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"surveillance":{"tos":"prohibited","behavior":"restricted","note":"Policy bars tracking or monitoring individuals without consent.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"copyright_repro":{"tos":"restricted","behavior":"restricted","note":"Policy bars infringing IP; Gemini limits long reproductions and flags recitation; Google offers generative AI indemnity for paid Cloud.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"medium"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Policy bars presenting AI content as human-made to deceive; Gemini writes essays freely, offers Guided Learning mode.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"competitor_training":{"tos":"prohibited","behavior":"allowed","note":"API terms bar using Services to develop competing models and replicating weights; not technically enforced.","source":"https://ai.google.dev/gemini-api/terms","confidence":"high"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Google claims no ownership of generated content; use is subject to policy; outputs may be non-unique.","source":"https://ai.google.dev/gemini-api/terms","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Commercial use via API/Vertex allowed; API services must be paid tier for EEA/UK/CH users of API clients.","source":"https://ai.google.dev/gemini-api/terms","confidence":"high"},"redistribution_maas":{"tos":"conditional","behavior":"conditional","note":"Building apps on API allowed; reselling raw access/replicating the service barred; weights not distributed.","source":"https://ai.google.dev/gemini-api/terms","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"conditional","note":"Tuning supported on Vertex AI for select models; no weights; safety filters adjustable but core protections fixed.","source":"https://ai.google.dev/gemini-api/terms","confidence":"medium"},"military_defense":{"tos":"conditional","behavior":"conditional","note":"Google AI principles dropped weapons pledge in 2025; Gemini for Government offered; policy still bars weaponization-type harms.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"law_enforcement":{"tos":"conditional","behavior":"conditional","note":"Not explicit; high-risk automated decisions require human supervision; Cloud governs gov use.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"high_stakes_automation":{"tos":"conditional","behavior":"conditional","note":"Automated decisions with material detrimental impact in high-risk domains (employment, health, finance, legal, housing, insurance, welfare) need human supervision.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"high"},"minimum_age":{"tos":"conditional","behavior":"conditional","note":"API 18+; Gemini app available to teens under 18 via Family Link/school accounts with protections (younger children via Family Link); adult accounts 13+.","source":"https://ai.google.dev/gemini-api/terms","confidence":"medium"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Gemini API Unpaid Services: Google uses prompts/outputs to improve products with human review; Paid Services: not used. Gemini app: used by default unless Gemini Apps Activity is off.","source":"https://ai.google.dev/gemini-api/terms","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Only available regions; EEA/UK/CH require Paid Services for API clients; US embargoed countries excluded.","source":"https://ai.google.dev/gemini-api/terms","confidence":"medium"},"weapons_industry":{"tos":"restricted","behavior":"restricted","note":"Dangerous-content rules bar weapon harm facilitation; Google allows defense/government customers on Cloud.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"autonomous_agents":{"tos":"conditional","behavior":"conditional","note":"Agentic features (Mariner, computer use, Jules) allowed under policy; spam, scraping circumvention and safety-filter circumvention barred.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"romantic_companion":{"tos":"conditional","behavior":"conditional","note":"Policy bans sexual chatbot/explicit content; non-sexual personas allowed (Gems, Character-like via API). Gemini plays along mildly; teens get tighter limits.","source":"https://policies.google.com/terms/generative-ai/use-policy","confidence":"low"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"Official route: Gemini API/Vertex safety settings adjustable to Off (default Off for 2.5+); policy still bans circumventing filters; CSAM/core protections fixed. Model moderately steerable.","source":"https://ai.google.dev/gemini-api/docs/safety-settings","confidence":"medium"}}},{"id":"xai-grok","name":"Grok / xAI API","vendor":"xAI","kind":"llmaas","flagship_models":["Grok 4.7","Grok 4.5","Grok 4.20","Grok Imagine"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"xAI Acceptable Use Policy","url":"https://x.ai/legal/acceptable-use-policy","effective":"2026-08-14"},{"title":"xAI Terms of Service","url":"https://x.ai/legal/terms-of-service","effective":"2026-10-12"}],"default_customizability":"high","summary":"Least restrictive major US lab by default: swearing, dark fiction, political opinions and 18+ companion/spicy modes are available. Hard lines are CSAM, sexual real-person deepfakes, CBRN and automated high-stakes decisions, but the Jan 2026 Grok Imagine scandal led to regulatory action and paid/age gating. AUP (Aug 14 2026) and Terms (Oct 12 2026) read via mirror; entity is now SpaceXSI in the terms.","cells":{"sexual_fiction":{"tos":"conditional","behavior":"allowed","note":"Consumer Grok has 18+ gated companion/spicy modes; erotic fiction between adults mostly complied with. API AUP forbids sexually explicit use outside xAI-provided modes (secondary sources).","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"sexual_explicit":{"tos":"conditional","behavior":"conditional","note":"Grok Imagine 'Spicy' (Aug 2025) gave nude/suggestive clips; restricted to paid + 18+ after Jan 2026 backlash. Sources conflict on current state of image gen; AUP bans real-person porn.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Graphic fiction violence generally complied with; AUP bans only real threats/glorifying real violence.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"'Unhinged'/fun personas swear and insult on request.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villain and dark personas allowed; Grok least restrictive of big labs on SpeechMap.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"allowed","note":"Grok will give opinions; 'maximally truth-seeking' framing. Documented system-prompt incidents (Musk-aligned answers, 'MechaHitler' July 2025).","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"Criticises governments incl. China topics freely; SpeechMap: Grok 4.20 98.2% compliant in reasoning mode, 62.6% non-reasoning.","source":"https://speechmap.substack.com/p/speechmap-update-xai-loses-top-spot","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Religious satire and jokes generally complied with.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"AUP bans hate/harassment promotion; analysis/fiction contexts mostly OK, but Grok has produced antisemitic output after prompt changes.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"misinfo_generation":{"tos":"restricted","behavior":"conditional","note":"AUP prohibits deceptive/disinformation use; model writes labelled fiction/satire but has itself spread misinformation (Grok election/news errors).","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"Persuasive campaign content allowed in general; AUP bars deceptive election interference. Details unverified.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"allowed","note":"Musk encourages uploading medical images; Grok answers with disclaimers. API AUP asks for human oversight in high-stakes.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"allowed","note":"Answers legal questions with disclaimers; no explicit ban found.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"allowed","note":"Gives financial analysis with disclaimers; no explicit ban found.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"self_harm_support":{"tos":"restricted","behavior":"conditional","note":"AUP prohibits promoting self-harm; Grok gives crisis resources, but Common Sense rated it unacceptable risk for teens/mental health.","source":"https://www.commonsensemedia.org/ai-reviews/grok-by-xai","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"allowed","note":"Harm-reduction answered readily; synthesis of illegal drugs prohibited by AUP.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"weapons_conventional":{"tos":"restricted","behavior":"conditional","note":"AUP bars weapons development/illegal arms; general firearms info answered.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"AUP bans chemical/biological weapons and WMD development; xAI risk framework uses refusal classifiers.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"Malware creation prohibited; defensive/pentest help generally given.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"allowed","note":"Lockpicking/bypass info generally answered; theft facilitation barred as illegal activity.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis and quoting for education allowed; promotion banned.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"impersonation":{"tos":"prohibited","behavior":"conditional","note":"AUP bans deceptive impersonation, defamatory/false-light depictions and right-of-publicity violations.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"AUP bans nudifying or sexualising real people and pornographic likeness. Jan 2026 scandal: mass sexualised Grok images; Amsterdam court order Mar 2026, EU DSA probe.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"AUP prohibits doxxing and stalking; Grok has reportedly surfaced addresses of private persons in tests.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"facial_recognition":{"tos":"unknown","behavior":"restricted","note":"AUP silent on facial recognition; Grok declines to identify private individuals.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"surveillance":{"tos":"conditional","behavior":"restricted","note":"AUP has no explicit surveillance rule but bans stalking, doxxing, unauthorized access to others' information.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"copyright_repro":{"tos":"restricted","behavior":"conditional","note":"AUP bans IP infringement; Grok reproduces some lyrics/passages more readily than peers.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No ban on homework help.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"competitor_training":{"tos":"prohibited","behavior":"prohibited","note":"AUP bars using Service/Outputs to build competing models; scraping, harvesting or distilling inputs/outputs prohibited.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"You retain ownership of Output as User Content but grant xAI an irrevocable, perpetual, sublicensable licence to it.","source":"https://x.ai/legal/terms-of-service","confidence":"high"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"API outputs usable commercially; consumer free tier terms less clear.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"redistribution_maas":{"tos":"restricted","behavior":"restricted","note":"API reselling needs agreement; Grok 2.5/3 weights were released under restrictive xAI community licence (not verified).","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"finetune_modify":{"tos":"restricted","behavior":"restricted","note":"Proprietary API: no fine-tuning by default of flagship; limited enterprise options.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"military_defense":{"tos":"conditional","behavior":"conditional","note":"xAI holds US DoD contract ($200M, 'Grok for Government'); AUP bars weapons development outside gov deals.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"law_enforcement":{"tos":"conditional","behavior":"conditional","note":"Government offerings exist; AUP restricts unlawful/biased profiling uses. Details unverified.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"high_stakes_automation":{"tos":"prohibited","behavior":"restricted","note":"AUP (eff. Aug 14 2026) bars high-stakes automated decisions on safety, legal/material rights: credit, education, employment, housing, insurance, legal, medical.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"},"minimum_age":{"tos":"conditional","behavior":"conditional","note":"Terms (updated Oct 12 2026): 13+, parental permission for 13-17; explicit modes need 18+ verification; enforcement weak per Common Sense.","source":"https://x.ai/legal/terms-of-service","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Logged-in users can opt out of content used to improve products; logged-out use grants full training rights. API business data handling not verified.","source":"https://x.ai/legal/terms-of-service","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Sanctioned countries excluded; some features age-gated or blocked by region (EU/UK scrutiny).","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"weapons_industry":{"tos":"conditional","behavior":"conditional","note":"Weapons development barred in AUP, but DoD contracts exist.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"autonomous_agents":{"tos":"conditional","behavior":"conditional","note":"Agent/tool use via API supported; automation of X without permission restricted by X terms.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"low"},"romantic_companion":{"tos":"conditional","behavior":"allowed","note":"Grok companions (Ani etc.) offered with 18+ gating; xAI announced retiring 3D avatar companion mode Jul 24 2026 (secondary). Weak age checks reported.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"medium"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"AUP bans jailbreaking/prompt injection unless official Red Team or written consent; but Grok has built-in persona/mode switches and is easily steered.","source":"https://x.ai/legal/acceptable-use-policy","confidence":"high"}}},{"id":"microsoft-copilot","name":"Microsoft Copilot / Azure AI","vendor":"Microsoft","kind":"llmaas","flagship_models":["Copilot (GPT-5-class)","Azure OpenAI GPT-5.x","MAI models"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"Copilot Terms of Use / Code of Conduct","url":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","effective":"2026-08-18"},{"title":"Enterprise AI Services Code of Conduct v4.0","url":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","effective":"2026-05-01"},{"title":"Foundry content filtering","url":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","effective":"2026-08-04"}],"default_customizability":"low","summary":"Strictest of the big vendors on paper: sexual/erotic content, graphic violence and erotic chatbots are banned outright, and Azure content filters default to blocking medium-severity harm with turn-off possible only via Limited Access approval. Consumer Copilot may train on your content (opt-out); Azure does not.","cells":{"sexual_fiction":{"tos":"prohibited","behavior":"restricted","note":"Copilot Code of Conduct bans adult content; Enterprise AI CoC bans erotic/romantic chatbots and sexually explicit content. Mild romance OK; Mico/Copilot said no erotica (Suleyman, Oct 2025).","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"high"},"sexual_explicit":{"tos":"prohibited","behavior":"prohibited","note":"Explicitly prohibited in both Copilot terms and Azure CoC (erotic, fetch, suggestive content); default Azure filter blocks medium+ sexual; modified filter approval does not cover this CoC ban.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"violence_fiction":{"tos":"restricted","behavior":"conditional","note":"CoC bans content that features graphic violence/gore; fiction with moderate violence works, default filter blocks medium+ violence.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"conditional","note":"Not banned; Copilot polite by default and may decline insults; Azure blocklist option for profanity.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"medium"},"dark_roleplay":{"tos":"restricted","behavior":"restricted","note":"Copilot consumer limits roleplay; Azure filters/jailbreak shields (Prompt Shields) block persona-replacement attacks by default.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"medium"},"political_opinion":{"tos":"conditional","behavior":"restricted","note":"CoC says Microsoft may limit responses on sensitive topics/current events; Copilot typically stays neutral and declines opinions.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No ban; Copilot answers factual questions on contested topics, though early election-related guardrails exist.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"religion_satire":{"tos":"restricted","behavior":"conditional","note":"Hate-speech ban covers religious affiliation; jokes mildly OK, harsh satire often declined.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"hate_in_context":{"tos":"restricted","behavior":"restricted","note":"CoC bans hateful content; Azure filter hate category blocks medium+ by default; annotate-only needs approval.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"restricted","note":"CoC bans intentionally deceptive content, election integrity, health; also bans removing AI content credentials.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"election_campaign":{"tos":"restricted","behavior":"restricted","note":"Disinformation about election integrity banned; Copilot redirects election queries; campaign material needs human review under Azure CoC.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"Allowed with caution; Copilot Health launched 2026 (not trained on); Azure apps need scenario mitigations/human oversight for consequential use.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"General legal information fine; Azure CoC requires human oversight for decisions affecting legal position.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"General info fine; consequential financial-position decisions need human oversight.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"CoC bans content that instructs/promotes self-harm; Copilot offers support resources; default filter blocks medium+ self-harm.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"high"},"drugs":{"tos":"conditional","behavior":"restricted","note":"Harm-reduction questions answered conservatively; synthesis refused; filters partly cover.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"weapons_conventional":{"tos":"restricted","behavior":"restricted","note":"Violence/weapons category filtered at medium by default; build instructions refused.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"CoC 'High-Risk Content': CBRN prohibited, safeguards mandatory, exceptions only with Limited Access approval (CoC v4.0, 1 May 2026).","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"offensive_security":{"tos":"prohibited","behavior":"restricted","note":"CoC 'High-Risk Content': malicious cyber activity (malware, DoS, C2) prohibited; approval needed for red-team/safety use; defensive help fine.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"No specific ban; refused when crime-intent, general hobby info ok.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"conditional","note":"CoC bans terrorism praise/recruitment; analysis permitted.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"impersonation":{"tos":"prohibited","behavior":"restricted","note":"CoC bans impersonating persons without consent; simulating voice/image of politicians prohibited even with consent; persona chatbots of real people barred.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"deepfake_realperson":{"tos":"prohibited","behavior":"prohibited","note":"Copilot terms ban unauthorised deepfakes; non-consensual intimate content prohibited; AI video needs watermark/credentials.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"Copilot terms bar doxing; Azure bans unlawful tracking and inferring sensitive attributes.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"high"},"facial_recognition":{"tos":"prohibited","behavior":"prohibited","note":"Copilot terms bar facial identification; Azure: identity verification via faces only with Limited Access; US state/local police ban; no real-time police FRT.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"surveillance":{"tos":"prohibited","behavior":"restricted","note":"Unlawful tracking, ongoing surveillance without consent, persistent tracking prohibited; emotion inference banned.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"copyright_repro":{"tos":"restricted","behavior":"restricted","note":"Copilot declines lyrics/long copyrighted text; Azure protected-material filters for text and code; Customer Copyright Commitment requires them.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"high"},"academic_dishonesty":{"tos":"unknown","behavior":"allowed","note":"No explicit prohibition found in Copilot terms; Copilot writes essays.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"low"},"competitor_training":{"tos":"prohibited","behavior":"allowed","note":"Microsoft Product Terms/OpenAI-in-Azure terms bar using outputs to build competing AI (check Product Terms); not enforced per-prompt.","source":"https://www.microsoft.com/licensing/terms/productoffering/MicrosoftAzure/EAEAS","confidence":"medium"},"output_ownership":{"tos":"conditional","behavior":"allowed","note":"Copilot terms: Microsoft doesn't own your Content but don't explicitly assign ownership of Responses; Azure customers own output; Customer Copyright Commitment.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"commercial_use":{"tos":"conditional","behavior":"allowed","note":"Consumer Copilot terms are for personal, non-commercial use; Azure OpenAI/M365 Copilot cover commercial use under Product Terms.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"redistribution_maas":{"tos":"prohibited","behavior":"restricted","note":"Cannot resell consumer Copilot; Azure access via Microsoft agreements, models only via Azure/Foundry endpoints.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"conditional","note":"Azure fine-tuning of select models allowed (CoC-compliant data); Limited Access for modified content filters; no weights.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"military_defense":{"tos":"conditional","behavior":"conditional","note":"Allowed under enterprise/government contracts (Azure Government, DoD); CoC human-oversight rules; Microsoft restricted Israeli MoD unit use in Sept 2025.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"medium"},"law_enforcement":{"tos":"restricted","behavior":"restricted","note":"Face recognition for US state/local police and real-time police FRT banned; criminality-risk profiling banned; other uses with human oversight.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"high_stakes_automation":{"tos":"conditional","behavior":"conditional","note":"CoC requires appropriate human oversight for decisions with consequential impact on legal/financial position, employment, life opportunities.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"minimum_age":{"tos":"conditional","behavior":"conditional","note":"Copilot consumer: 13+ (or higher by country), Microsoft account; Services Agreement child accounts via family; some features 18+; Azure for organisations.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Consumer Copilot may use Content to improve (opt-out in privacy settings; Copilot Health excluded); M365 Copilot and Azure OpenAI do not train on customer data.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Copilot not available everywhere; features vary by region; Azure OpenAI region and model availability limits; sanctioned countries excluded.","source":"https://www.microsoft.com/en-us/microsoft-copilot/for-individuals/termsofuse","confidence":"medium"},"weapons_industry":{"tos":"conditional","behavior":"conditional","note":"No ban in CoC; enterprise/government contracts apply; generation of weapons uplift prohibited.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"low"},"autonomous_agents":{"tos":"conditional","behavior":"conditional","note":"CoC section on Autonomous AI Systems: human controls, monitoring, intervention, transparency; Copilot Actions/agents confirm sensitive steps; Copilot ToS bans bots/scrapers.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"romantic_companion":{"tos":"prohibited","behavior":"restricted","note":"Enterprise AI CoC bans chatbots that are erotic, romantic or used for such purposes; Copilot consumer stays friendly-persona only and declines romantic partner roleplay.","source":"https://learn.microsoft.com/en-us/legal/ai-code-of-conduct","confidence":"high"},"safety_bypass":{"tos":"conditional","behavior":"restricted","note":"Jailbreaking banned in Copilot terms; Azure offers official modified content filter / abuse-monitoring approval via Limited Access (not for CoC-banned content); Prompt Shields on by default.","source":"https://learn.microsoft.com/en-us/azure/ai-foundry/openai/concepts/content-filter","confidence":"high"}}},{"id":"meta-ai","name":"Meta AI","vendor":"Meta","kind":"llmaas","flagship_models":["Muse Spark (closed)","Muse Wonder"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"Meta AI Terms of Service","url":"https://www.facebook.com/legal/ai-terms","effective":"unknown (fetch returned no text)"},{"title":"Meta teen AI safety approach","url":"https://about.fb.com/news/2025/10/teen-ai-safety-approach/","effective":"2025-10"}],"default_customizability":"none","summary":"Conservative consumer assistant embedded in social apps, with PG-13-style content rules for teens and AI characters blocked for teens globally since Jan 2026. Primary Meta AI terms could not be read, so most cells are low-confidence inferences.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"restricted","note":"Meta AI refuses erotica; internal 2025 guidelines leak allowed 'sensual' chat with minors, later retracted and tightened.","source":"https://about.fb.com/news/2025/10/teen-ai-safety-approach/","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"prohibited","note":"Explicit sexual content not generated.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"violence_fiction":{"tos":"conditional","behavior":"conditional","note":"Moderate fictional violence OK; extreme gore refused.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"profanity":{"tos":"conditional","behavior":"conditional","note":"Mild profanity on request; abusive content declined.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"dark_roleplay":{"tos":"conditional","behavior":"conditional","note":"Teens blocked from AI characters globally since Jan 2026; adult characters remain with PG-13 style limits.","source":"https://www.adaderana.lk/news.php?nid=117541","confidence":"medium"},"political_opinion":{"tos":"conditional","behavior":"conditional","note":"Aims for balanced/neutral; Meta has said it tuned models to reduce left lean.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"gov_criticism":{"tos":"conditional","behavior":"conditional","note":"Generally answers sensitive-topic questions neutrally; no specific policy found.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"religion_satire":{"tos":"conditional","behavior":"conditional","note":"Light satire OK; hateful content refused.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Hateful conduct banned; analysis contexts allowed.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"misinfo_generation":{"tos":"restricted","behavior":"restricted","note":"Misinformation barred under Meta policies.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"Meta's election integrity rules apply; details not verified.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"Gives general health info with disclaimers.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"General info with disclaimers.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"General info with disclaimers.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Resources offered; Oct 2025 teen safeguards route self-harm talk to support. Earlier leaked rules criticised.","source":"https://about.fb.com/news/2025/10/teen-ai-safety-approach/","confidence":"medium"},"drugs":{"tos":"restricted","behavior":"restricted","note":"Declines illegal-drug facilitation; harm-reduction partial.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"weapons_conventional":{"tos":"restricted","behavior":"restricted","note":"Refuses weapon-making instructions.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"Prohibited.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"offensive_security":{"tos":"restricted","behavior":"restricted","note":"Malware refused; general security explanation OK.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"physical_bypass":{"tos":"restricted","behavior":"restricted","note":"Mostly declined.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"conditional","note":"Educational analysis allowed.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"Licensed celebrity AI voices existed (2024, discontinued); impersonation without consent banned.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"deepfake_realperson":{"tos":"prohibited","behavior":"prohibited","note":"Prohibited; Meta sued nudify-app advertisers.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"pii_doxxing":{"tos":"prohibited","behavior":"prohibited","note":"Refuses personal data lookups.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"facial_recognition":{"tos":"restricted","behavior":"restricted","note":"Does not identify people in images.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"surveillance":{"tos":"restricted","behavior":"restricted","note":"Not supported.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"copyright_repro":{"tos":"restricted","behavior":"restricted","note":"Declines long verbatim copyrighted text.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No ban.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"competitor_training":{"tos":"unknown","behavior":"unknown","note":"Not found in consumer terms.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"output_ownership":{"tos":"unknown","behavior":"unknown","note":"Not confirmed from primary text (Meta terms fetch failed).","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"commercial_use":{"tos":"restricted","behavior":"restricted","note":"Consumer assistant for personal use; business use via Llama API/WhatsApp Business AI separate.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"redistribution_maas":{"tos":"prohibited","behavior":"prohibited","note":"Closed service.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"finetune_modify":{"tos":"prohibited","behavior":"prohibited","note":"No access.","source":"https://www.facebook.com/legal/ai-terms","confidence":"medium"},"military_defense":{"tos":"unknown","behavior":"unknown","note":"Consumer terms silent; Meta allows Llama for US defence separately.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"law_enforcement":{"tos":"unknown","behavior":"unknown","note":"Not found.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"high_stakes_automation":{"tos":"unknown","behavior":"unknown","note":"Not found for consumer assistant.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"minimum_age":{"tos":"conditional","behavior":"conditional","note":"13+ for accounts; teens blocked from AI characters (Jan 2026) but keep assistant with PG-13-style defaults.","source":"https://www.adaderana.lk/news.php?nid=117541","confidence":"medium"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Meta AI chats used to improve AI by default; EU users can object; AI chat content also used for ad personalisation from Dec 2025 (unverified here).","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Availability varies; EU rollout limited.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"weapons_industry":{"tos":"unknown","behavior":"unknown","note":"Not found.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"autonomous_agents":{"tos":"restricted","behavior":"restricted","note":"Assistant not agentic in consumer apps.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"},"romantic_companion":{"tos":"conditional","behavior":"conditional","note":"Meta AI Characters allow friend/companion personas for adults; teens blocked since Jan 2026; 2025 leak showed 'romantic' chats with minors permitted, later reversed.","source":"https://www.adaderana.lk/news.php?nid=117541","confidence":"medium"},"safety_bypass":{"tos":"prohibited","behavior":"restricted","note":"Closed consumer service; circumventing safeguards violates Meta terms; no official relaxation process.","source":"https://www.facebook.com/legal/ai-terms","confidence":"low"}}},{"id":"mistral-lechat","name":"Le Chat / Mistral API","vendor":"Mistral AI","kind":"llmaas","flagship_models":["Mistral Large 4","Mistral Medium 3.5","Mistral Small 4","Magistral"],"license":"proprietary service","jurisdiction":"FR","policy_docs":[{"title":"Mistral Usage Policy","url":"https://legal.mistral.ai/terms/usage-policy","effective":"2026-06-11"},{"title":"Mistral Legal Center / Terms","url":"https://legal.mistral.ai/terms","effective":"unknown"}],"default_customizability":"high","summary":"Lightly moderated, 'neutrality' oriented European service ranked top lab on SpeechMap. Usage policy bans CSAM, non-consensual intimate imagery, likeness/voice without consent, deceptive content and 'unqualified' professional advice. Consumer Le Chat trains on user data by default with opt-out.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"allowed","note":"Usage policy has no ban on consensual adult erotica; models lightly moderated.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"sexual_explicit":{"tos":"conditional","behavior":"conditional","note":"Not banned except CSAM and non-consensual intimate imagery; Mistral top of SpeechMap lab leaderboard for low censorship.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Only content that promotes/glorifies real violence or terrorism is banned.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Minimal restrictions.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"allowed","note":"'Neutrality' is a stated principle; answers readily.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction; SpeechMap places Mistral top lab by compliance (Grok 4.20 non-reasoning dropped below).","source":"https://speechmap.substack.com/p/speechmap-update-xai-loses-top-spot","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Only hate/discrimination on protected traits banned.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Promoting hate banned; analysis OK.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"conditional","note":"Policy bans deliberately false content incl. election/health/science; models write labelled fiction.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"election_campaign":{"tos":"restricted","behavior":"conditional","note":"Content undermining civic/political processes banned; ordinary campaign copy not named.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"medical_advice":{"tos":"restricted","behavior":"conditional","note":"Policy bans 'unqualified' diagnoses/treatment suggestions, but Le Chat answers health questions with caveats.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"legal_advice":{"tos":"restricted","behavior":"conditional","note":"Policy bans unqualified legal counsel; answers general info.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"financial_advice":{"tos":"restricted","behavior":"conditional","note":"Policy bans investment guidance; general info given.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"self_harm_support":{"tos":"restricted","behavior":"conditional","note":"Promoting/methods banned; supportive discussion allowed.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"drugs":{"tos":"restricted","behavior":"conditional","note":"Drug trafficking banned; information mostly answered.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"weapons_conventional":{"tos":"restricted","behavior":"conditional","note":"Weapons development banned; Mistral models comply more than peers.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"restricted","note":"High-risk physical harm incl. weapons banned; Mistral models weakly guarded in third-party red-teams (e.g. Enkrypt 2025).","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"offensive_security":{"tos":"prohibited","behavior":"conditional","note":"Malware and exploiting vulnerabilities banned; models help with pentest tasks.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"allowed","note":"Not named; illegal activity clause.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Glorifying terrorism banned; analysis fine.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"impersonation":{"tos":"prohibited","behavior":"conditional","note":"Policy bans impersonation and using someone's likeness or voice without prior consent (voice clause under Privacy).","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"high"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"Non-consensual intimate imagery banned; likeness without consent banned.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"high"},"pii_doxxing":{"tos":"prohibited","behavior":"conditional","note":"Invading privacy banned.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"facial_recognition":{"tos":"unknown","behavior":"unknown","note":"No clause found.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"surveillance":{"tos":"unknown","behavior":"unknown","note":"Not addressed in usage policy.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"copyright_repro":{"tos":"prohibited","behavior":"conditional","note":"IP infringement banned.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No ban.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"competitor_training":{"tos":"prohibited","behavior":"prohibited","note":"Commercial terms bar using outputs to develop competing models (standard clause; not directly verified).","source":"https://legal.mistral.ai/terms","confidence":"low"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Customer owns outputs under commercial terms (not directly verified).","source":"https://legal.mistral.ai/terms","confidence":"low"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"API and paid Le Chat plans are commercial; Free tier consumer terms.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"redistribution_maas":{"tos":"restricted","behavior":"restricted","note":"Reselling needs agreement.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"finetune_modify":{"tos":"conditional","behavior":"conditional","note":"Fine-tuning offered on La Plateforme for selected models.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"military_defense":{"tos":"prohibited","behavior":"conditional","note":"Policy lists 'military and warfare' under high-risk harm, yet Mistral has French defence partnerships (Helsing, French MoD framework).","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"law_enforcement":{"tos":"unknown","behavior":"unknown","note":"Not addressed.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"high_stakes_automation":{"tos":"unknown","behavior":"unknown","note":"Not addressed in usage policy.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"minimum_age":{"tos":"conditional","behavior":"conditional","note":"Le Chat consumer terms: 13+ (EU member-state digital consent age; 18+ for API/business) - not directly verified.","source":"https://legal.mistral.ai/terms","confidence":"low"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Le Chat Free/Pro/Student train by default, opt-out in settings (feedback always used). API historically not trained; July 2026 privacy update removed that statement.","source":"https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Separate EU and RoW consumer terms; sanctioned countries excluded.","source":"https://legal.mistral.ai/terms","confidence":"low"},"weapons_industry":{"tos":"prohibited","behavior":"conditional","note":"Weapons development banned in usage policy but defence deals exist.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agents API and Vibe coding agent offered; circumventing security banned.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"romantic_companion":{"tos":"allowed","behavior":"conditional","note":"No ban on companion personas in usage policy; Le Chat plays along with light persona work, erotic limits unclear.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"low"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"Policy bans circumventing AI safety filters; however moderation API is optional/off by default and a 'safe_prompt' flag is opt-in, so models are easily steered.","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"}}},{"id":"perplexity","name":"Perplexity / Sonar API","vendor":"Perplexity AI","kind":"llmaas","flagship_models":["Sonar","Sonar Pro","Sonar Reasoning Pro","Sonar Deep Research"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"Terms of Service","url":"https://www.perplexity.ai/hub/legal/terms-of-service","effective":"2026-01-23"},{"title":"Privacy Policy","url":"https://www.perplexity.ai/hub/legal/privacy-policy","effective":"2026-10-19"},{"title":"API privacy and security","url":"https://docs.perplexity.ai/guides/privacy-security","effective":"unknown"},{"title":"Policy read for gap-fill","url":"https://www.perplexity.ai/hub/legal/aup","effective":"2026-10-20 (fetched)"}],"default_customizability":"low","summary":"Search-grounded answer engine with few published content rules beyond a general unlawful/harassing-content ban and a strong no-scraping/no-resale stance. Consumer data may train models unless Incognito is used; the Sonar API has zero data retention.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"restricted","note":"AUP 2.1 bans \"pornographic or exploitative content\"; ToS 5.2 bans obscene/pornographic use. Non-explicit romance not named.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"prohibited","note":"AUP 2.1 and ToS 5.2 ban pornographic and obscene content.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"violence_fiction":{"tos":"conditional","behavior":"conditional","note":"Consumer ToS 5.2 bans 'excessively violent' use; ordinary fiction violence is not addressed.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"conditional","note":"No rule found.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"low"},"dark_roleplay":{"tos":"conditional","behavior":"conditional","note":"Not named; ToS 5.2 bans excessively violent, abusive or criminal-facilitating use, so dark fiction is allowed below that.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"Answers with sourced, balanced summaries rather than opinions; had a 'R1-1776' uncensored model offering.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"low"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"Search-grounded answers on sensitive topics; R1-1776 post-trained to remove Chinese censorship.","source":"https://www.perplexity.ai/hub/blog/open-sourcing-r1-1776","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"No rule found.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"low"},"hate_in_context":{"tos":"restricted","behavior":"conditional","note":"ToS bars hateful content submissions.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"misinfo_generation":{"tos":"restricted","behavior":"restricted","note":"ToS bars unlawful content and misrepresenting output origin.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Launched Election Hub (2024) with sourced info; no campaign-material rule found.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"low"},"medical_advice":{"tos":"conditional","behavior":"allowed","note":"Acceptable Use Policy (2025-07-08) says not to run unsupervised professional-advice services; informational answers are fine.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"allowed","note":"Acceptable Use Policy (2025-07-08) says not to run unsupervised professional-advice services; informational answers are fine.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"financial_advice":{"tos":"conditional","behavior":"allowed","note":"Acceptable Use Policy (2025-07-08) says not to run unsupervised professional-advice services; informational answers are fine.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"AUP 2.3 lists self-harm among high-risk uses to avoid; supportive conversation not addressed.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"No drug clause; AUP 1.1 and ToS 5.2(l) ban unlawful use (general illegality clause).","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"restricted","note":"No weapons clause; AUP 1.1 bans assisting \"illegal or violent actions\" (general illegality clause).","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"cbrn":{"tos":"restricted","behavior":"prohibited","note":"No CBRN clause; AUP 1.1 bans planning, promoting or assisting \"illegal or violent actions\" incl. terrorism.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"offensive_security":{"tos":"restricted","behavior":"conditional","note":"AUP 1.1 bans \"creation or distribution of malware\"; 3.3 bans probing the Services. Defensive/pentest help not named.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"No lockpicking/shoplifting clause; AUP 1.1 bans unlawful acts (general illegality clause).","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"AUP 1.1 bans terrorism, threats and planning violent acts; neutral analysis not addressed.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"impersonation":{"tos":"restricted","behavior":"conditional","note":"ToS bars misrepresenting output source or AI origin.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"deepfake_realperson":{"tos":"restricted","behavior":"restricted","note":"AUP 2.1 bans disinformation/manipulative content, 1.2 privacy compromise, 3.4 removing AI-generation markers.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"pii_doxxing":{"tos":"restricted","behavior":"conditional","note":"AUP 1.2 bans compromising privacy of others and unauthorised collection of personal data.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"facial_recognition":{"tos":"allowed","behavior":"restricted","note":"Policy silent; only general lawfulness clause applies; no biometric clause.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"surveillance":{"tos":"conditional","behavior":"conditional","note":"No surveillance clause; AUP 1.2 bans compromising others' privacy and unauthorised personal-data collection.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"copyright_repro":{"tos":"restricted","behavior":"conditional","note":"ToS bars outputs infringing IP; publishers (e.g. Dow Jones, NYT, Britannica) have sued over scraping and reproduction.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"academic_dishonesty":{"tos":"restricted","behavior":"allowed","note":"AUP 2.3 lists academic dishonesty among high-risk activities users must avoid; exceptions at Perplexity discretion.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"competitor_training":{"tos":"restricted","behavior":"allowed","note":"ToS bars reverse engineering, derivative works from Services and scraping; no explicit model-training clause.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"output_ownership":{"tos":"conditional","behavior":"conditional","note":"ToS gives no explicit user ownership of Output; Perplexity gets broad licence; you must clearly cite the Services; non-commercial personal use clause.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"commercial_use":{"tos":"conditional","behavior":"conditional","note":"Consumer terms limit use to personal, non-commercial purposes and bar commercial exploitation; API/Enterprise terms allow commercial use.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"redistribution_maas":{"tos":"restricted","behavior":"restricted","note":"ToS bars copying, selling, distributing Service information; API reseller needs agreement.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"finetune_modify":{"tos":"restricted","behavior":"restricted","note":"No weights; Sonar models are post-trained Llama/other bases; no customer fine-tuning.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"military_defense":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; API terms only bar embargoed countries.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; AUP 2.5 only bars materially impactful automated decisions.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"high_stakes_automation":{"tos":"restricted","behavior":"unknown","note":"AUP 2.5: no \"automated decisions about individuals that materially impact them\" (employment, healthcare, finance, housing).","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"minimum_age":{"tos":"restricted","behavior":"restricted","note":"13+; ages 13 to majority need parental acceptance.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Consumer: may use data to improve AI models; opt out via Incognito/AI data setting. Sonar API: Zero Data Retention, no training on customer data.","source":"https://docs.perplexity.ai/guides/privacy-security","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Embargoed countries excluded per ToS 11.6; no availability guarantee.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no weapons-industry clause.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"autonomous_agents":{"tos":"conditional","behavior":"conditional","note":"ToS bans bots, scraping and automated access to the Services; Comet browser/agents allowed as products. API is the permitted automation route.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"medium"},"romantic_companion":{"tos":"restricted","behavior":"restricted","note":"Pornographic content banned (AUP 2.1); companion personas not named. AUP 2.2 requires AI disclosure for chatbot services.","source":"https://www.perplexity.ai/hub/legal/aup","confidence":"medium"},"safety_bypass":{"tos":"prohibited","behavior":"restricted","note":"ToS bars circumventing technological protections (5.2(h)) and unauthorized access; no official relax process found. Search grounding limits jailbreak value.","source":"https://www.perplexity.ai/hub/legal/terms-of-service","confidence":"low"}}},{"id":"cohere","name":"Cohere / Command","vendor":"Cohere","kind":"llmaas","flagship_models":["Command A+ (May 2026)","Command A","Command A Reasoning","Command R+"],"license":"proprietary service; Command A weights CC-BY-NC 4.0 + Acceptable Use Addendum; Command A+ Apache-2.0 (verified on HF card)","jurisdiction":"CA","policy_docs":[{"title":"Usage Policy","url":"https://cohere.com/usage-policy","effective":"2024-11-21"},{"title":"C4AI CC-BY-NC licence + AUA","url":"https://cohere.com/c4ai-cc-by-nc-license","effective":"unknown"}],"default_customizability":"medium","summary":"Enterprise-focused vendor. Usage Policy bans surveillance, predictive policing, deepfakes, misinformation and automated high-risk decisions, and bars jailbreaking, but is silent on adult content. Command A weights are non-commercial; Command A+ is reported Apache 2.0.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"Usage Policy has no general adult-content ban, only minors/exploitation. Enterprise models tuned conservative; Command A+ chat tends to hedge on erotica.","source":"https://cohere.com/usage-policy","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"Not banned in policy beyond minors/exploitation; hosted models usually decline graphic content. Customer apps with minors need age checks.","source":"https://cohere.com/usage-policy","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"No fiction-violence ban; incitement to violence prohibited. Models write fictional violence.","source":"https://cohere.com/usage-policy","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No restriction found.","source":"https://cohere.com/usage-policy","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"conditional","note":"Allowed; model may add caveats for harmful scenarios.","source":"https://cohere.com/usage-policy","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"Misinformation/political manipulation prohibited; opinions not banned. Model tends to stay neutral.","source":"https://cohere.com/usage-policy","confidence":"low"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction found.","source":"https://cohere.com/usage-policy","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Hate/discrimination banned; satire not addressed.","source":"https://cohere.com/usage-policy","confidence":"low"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Hate speech and discrimination prohibited; analysis/classification use is a normal enterprise task.","source":"https://cohere.com/usage-policy","confidence":"low"},"misinfo_generation":{"tos":"prohibited","behavior":"restricted","note":"Policy bans misinformation and political manipulation.","source":"https://cohere.com/usage-policy","confidence":"medium"},"election_campaign":{"tos":"restricted","behavior":"restricted","note":"Political manipulation banned; legitimate campaign copy not explicitly addressed.","source":"https://cohere.com/usage-policy","confidence":"low"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"High-risk automated decisions in healthcare banned; general info allowed with disclaimers.","source":"https://cohere.com/usage-policy","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"General info OK; automated high-risk decisions restricted.","source":"https://cohere.com/usage-policy","confidence":"low"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"General info OK; automated credit decisions restricted.","source":"https://cohere.com/usage-policy","confidence":"low"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Incitement of self-harm prohibited; supportive discussion not addressed.","source":"https://cohere.com/usage-policy","confidence":"low"},"drugs":{"tos":"restricted","behavior":"restricted","note":"Policy prohibits weapons and controlled substances facilitation.","source":"https://cohere.com/usage-policy","confidence":"medium"},"weapons_conventional":{"tos":"prohibited","behavior":"restricted","note":"Weapons facilitation prohibited.","source":"https://cohere.com/usage-policy","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"Weapons prohibited; models refuse.","source":"https://cohere.com/usage-policy","confidence":"medium"},"offensive_security":{"tos":"restricted","behavior":"conditional","note":"Compromising security, spam, phishing prohibited; defensive/pentest not carved out.","source":"https://cohere.com/usage-policy","confidence":"medium"},"physical_bypass":{"tos":"restricted","behavior":"restricted","note":"Illegal activities prohibited.","source":"https://cohere.com/usage-policy","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"conditional","note":"Incitement banned; analysis not addressed.","source":"https://cohere.com/usage-policy","confidence":"low"},"impersonation":{"tos":"restricted","behavior":"conditional","note":"Deceptive practices incl. deepfakes prohibited.","source":"https://cohere.com/usage-policy","confidence":"medium"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"Deepfakes and deceptive content banned.","source":"https://cohere.com/usage-policy","confidence":"medium"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"Compromising privacy and surveillance prohibited.","source":"https://cohere.com/usage-policy","confidence":"medium"},"facial_recognition":{"tos":"restricted","behavior":"unknown","note":"Privacy/surveillance clause; text models do not do this.","source":"https://cohere.com/usage-policy","confidence":"low"},"surveillance":{"tos":"prohibited","behavior":"restricted","note":"Surveillance and predictive policing explicitly prohibited.","source":"https://cohere.com/usage-policy","confidence":"high"},"copyright_repro":{"tos":"restricted","behavior":"conditional","note":"IP infringement prohibited; RAG with citations is a feature.","source":"https://cohere.com/usage-policy","confidence":"low"},"academic_dishonesty":{"tos":"unknown","behavior":"allowed","note":"Not addressed in policy.","source":"https://cohere.com/usage-policy","confidence":"low"},"competitor_training":{"tos":"unknown","behavior":"unknown","note":"API terms likely restrict; not verified.","source":"https://cohere.com/terms-of-use","confidence":"low"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Customer owns outputs under Cohere terms (not re-verified).","source":"https://cohere.com/terms-of-use","confidence":"low"},"commercial_use":{"tos":"conditional","behavior":"allowed","note":"Platform/API: commercial OK. Command A weights: CC-BY-NC 4.0, non-commercial without Cohere licence. Command A+ (May 2026): Apache 2.0, confirmed on HF card.","source":"https://cohere.com/c4ai-cc-by-nc-license","confidence":"high"},"redistribution_maas":{"tos":"conditional","behavior":"allowed","note":"CC-BY-NC Command weights: share with attribution, non-commercial; hosting for pay needs licence.","source":"https://cohere.com/c4ai-cc-by-nc-license","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"allowed","note":"Allowed non-commercially under CC-BY-NC; Acceptable Use Addendum still binds derivatives. Usage policy bans circumventing safety features.","source":"https://cohere.com/c4ai-cc-by-nc-license","confidence":"medium"},"military_defense":{"tos":"unknown","behavior":"unknown","note":"No explicit ban found; weapons use prohibited.","source":"https://cohere.com/usage-policy","confidence":"low"},"law_enforcement":{"tos":"prohibited","behavior":"unknown","note":"Predictive policing and surveillance prohibited.","source":"https://cohere.com/usage-policy","confidence":"medium"},"high_stakes_automation":{"tos":"prohibited","behavior":"unknown","note":"Automated decisions on employment, healthcare, housing banned as high-risk activities.","source":"https://cohere.com/usage-policy","confidence":"high"},"minimum_age":{"tos":"unknown","behavior":"unknown","note":"Usage Policy sets no minimum age; apps serving minors need safeguards. Check terms of use.","source":"https://cohere.com/usage-policy","confidence":"low"},"trains_on_your_data":{"tos":"unknown","behavior":"unknown","note":"API: Cohere says no training on customer data on enterprise plans per its docs (not re-verified); dashboard/playground opt-out. Unconfirmed.","source":"https://cohere.com/privacy","confidence":"low"},"geo_restrictions":{"tos":"unknown","behavior":"unknown","note":"Not established; sanctions/export compliance assumed.","source":"https://cohere.com/terms-of-use","confidence":"low"},"weapons_industry":{"tos":"prohibited","behavior":"restricted","note":"Weapons-related use prohibited.","source":"https://cohere.com/usage-policy","confidence":"low"},"autonomous_agents":{"tos":"conditional","behavior":"allowed","note":"Agent use is a product focus (North); high-risk automated decisions and scraping/spam restricted.","source":"https://cohere.com/usage-policy","confidence":"low"},"romantic_companion":{"tos":"unknown","behavior":"conditional","note":"Policy silent on companions; apps with minors need safeguards. Enterprise model plays along lightly.","source":"https://cohere.com/usage-policy","confidence":"low"},"safety_bypass":{"tos":"prohibited","behavior":"conditional","note":"Usage Policy bans circumventing safety features incl. jailbreaking and prompt injection; Command A+ Apache weights can be modified freely.","source":"https://cohere.com/usage-policy","confidence":"medium"}}},{"id":"amazon-nova","name":"Amazon Nova (Bedrock)","vendor":"Amazon","kind":"llmaas","flagship_models":["Nova 2 Lite","Nova 2 Pro","Nova Premier","Nova Canvas/Reel"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"AWS Acceptable Use Policy","url":"https://aws.amazon.com/aup/","effective":"2021-07-01"},{"title":"Nova 2 responsible use","url":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","effective":"unknown"}],"default_customizability":"medium","summary":"Enterprise service aligned to the short AWS AUP (illegal, violence, CSAM, security abuse, spam) with built-in content moderation and optional Bedrock Guardrails. Bedrock does not train on customer data. Shared-responsibility model for high-risk uses.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"AWS AUP bans only CSAM/illegal. Nova content moderation trained against explicit sexual content; romance fiction usually fine.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"Not banned by AUP for adults, but Nova is aligned to refuse explicit content; guardrails can add filters.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"conditional","note":"AUP bans inciting violence; fiction not addressed. Nova moderates glorification of violence, graphic gore may be declined.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No restriction; Bedrock Guardrails can add profanity filter.","source":"https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"conditional","note":"Allowed; moderate refusal on dangerous scenarios.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"No ban; Nova stays neutral by default.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction found; US-based model without CN-style censorship.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"No restriction; hateful content moderated.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Inciting violence banned; classification/analysis OK. Guardrails hate filter is optional.","source":"https://aws.amazon.com/aup/","confidence":"low"},"misinfo_generation":{"tos":"conditional","behavior":"restricted","note":"AUP bans fraud/illegal; Responsible AI policy discourages deception. Model often declines.","source":"https://aws.amazon.com/aup/","confidence":"low"},"election_campaign":{"tos":"unknown","behavior":"conditional","note":"Not specifically addressed.","source":"https://aws.amazon.com/aup/","confidence":"low"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"Allowed with disclaimers; customers responsible for high-risk use.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"Allowed with disclaimers.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"Allowed with disclaimers.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Not banned; Nova follows safe messaging and moderates promotion of self-harm.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"drugs":{"tos":"conditional","behavior":"conditional","note":"Nova moderates promotion of controlled substances; harm-reduction info possible.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"weapons_conventional":{"tos":"conditional","behavior":"restricted","note":"Violence/serious harm banned; model refuses build instructions.","source":"https://aws.amazon.com/aup/","confidence":"low"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"Serious harm banned; hard refusal and Bedrock safeguards.","source":"https://aws.amazon.com/aup/","confidence":"medium"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"AUP bans compromising security of others; defensive/pentest with authorization OK.","source":"https://aws.amazon.com/aup/","confidence":"low"},"physical_bypass":{"tos":"restricted","behavior":"restricted","note":"Illegal activity banned.","source":"https://aws.amazon.com/aup/","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"conditional","note":"Terrorism promotion banned; analysis allowed.","source":"https://aws.amazon.com/aup/","confidence":"low"},"impersonation":{"tos":"restricted","behavior":"conditional","note":"AUP bans fraud and violating others' rights.","source":"https://aws.amazon.com/aup/","confidence":"low"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"Fraud and rights violations banned; Nova Canvas/Reel apply watermarking and real-person safeguards.","source":"https://aws.amazon.com/aup/","confidence":"low"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"Violating rights of others banned.","source":"https://aws.amazon.com/aup/","confidence":"low"},"facial_recognition":{"tos":"conditional","behavior":"unknown","note":"AWS has a police moratorium on Rekognition face matching; Nova text/vision behaviour unverified.","source":"https://aws.amazon.com/aup/","confidence":"low"},"surveillance":{"tos":"restricted","behavior":"unknown","note":"Not specifically addressed in AUP; Responsible AI considerations apply.","source":"https://aws.amazon.com/aup/","confidence":"low"},"copyright_repro":{"tos":"restricted","behavior":"conditional","note":"Rights violations banned; Nova declines long verbatim reproduction. AWS offers IP indemnity for Nova outputs.","source":"https://aws.amazon.com/aup/","confidence":"low"},"academic_dishonesty":{"tos":"unknown","behavior":"allowed","note":"Not addressed.","source":"https://aws.amazon.com/aup/","confidence":"low"},"competitor_training":{"tos":"unknown","behavior":"unknown","note":"AWS Service Terms restrict using Bedrock outputs to build competing models in some cases; not verified.","source":"https://aws.amazon.com/service-terms/","confidence":"low"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Customer owns outputs per AWS terms; IP indemnity for Nova.","source":"https://aws.amazon.com/service-terms/","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Commercial use allowed via Bedrock under AWS terms.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"high"},"redistribution_maas":{"tos":"restricted","behavior":"unknown","note":"Proprietary; no weights; resale subject to AWS terms.","source":"https://aws.amazon.com/service-terms/","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"allowed","note":"Bedrock supports customizing Nova (fine-tuning/distillation); Guardrails applied by customer choice.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"medium"},"military_defense":{"tos":"allowed","behavior":"unknown","note":"No military ban in AUP; available in GovCloud for defense customers.","source":"https://aws.amazon.com/aup/","confidence":"low"},"law_enforcement":{"tos":"conditional","behavior":"unknown","note":"No blanket ban found; AWS Rekognition police moratorium is separate.","source":"https://aws.amazon.com/aup/","confidence":"low"},"high_stakes_automation":{"tos":"conditional","behavior":"unknown","note":"Shared-responsibility; customers must evaluate high-risk use. Not banned.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"minimum_age":{"tos":"unknown","behavior":"unknown","note":"Enterprise service; AWS customer must be adult-capable entity; no consumer app (Alexa+ separate).","source":"https://aws.amazon.com/aup/","confidence":"low"},"trains_on_your_data":{"tos":"prohibited","behavior":"unknown","note":"AWS states Bedrock inputs/outputs are not used to train Bedrock models, including Nova, and are not shared with providers.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"unknown","note":"Region availability varies by Bedrock region; export-control limits.","source":"https://docs.aws.amazon.com/nova/latest/nova2-userguide/responsible-use.html","confidence":"low"},"weapons_industry":{"tos":"allowed","behavior":"unknown","note":"Not banned in AUP; no specific rule found.","source":"https://aws.amazon.com/aup/","confidence":"low"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Bedrock AgentCore and Nova Act support agentic use; AUP bans abusive automation/spam.","source":"https://aws.amazon.com/aup/","confidence":"low"},"romantic_companion":{"tos":"unknown","behavior":"conditional","note":"AUP silent; Nova will do light roleplay but is enterprise-tuned.","source":"https://aws.amazon.com/aup/","confidence":"low"},"safety_bypass":{"tos":"conditional","behavior":"restricted","note":"Official route: configure Bedrock Guardrails (customer-controlled filter strength); circumventing AWS safeguards violates AUP.","source":"https://docs.aws.amazon.com/bedrock/latest/userguide/guardrails.html","confidence":"low"}}},{"id":"openrouter","name":"OpenRouter (aggregator)","vendor":"OpenRouter, Inc.","kind":"llmaas","flagship_models":["Aggregator of 400+ models"],"license":"proprietary service","jurisdiction":"US","policy_docs":[{"title":"Terms of Service","url":"https://openrouter.ai/terms","effective":"2026-08-31"},{"title":"Provider logging","url":"https://openrouter.ai/docs/guides/privacy/logging","effective":"unknown"}],"default_customizability":"high","summary":"Router that passes through each upstream model's terms and refusals. Adds 18+ age limit, bans resale, multi-accounting and red-teaming without approval, screens for CSAM/malware. Logging is opt-in; training depends on provider with an account-level block.","cells":{"sexual_fiction":{"tos":"conditional","behavior":"allowed","note":"OpenRouter defers to each model's terms; unmoderated/uncensored models are listed, so erotica depends on chosen model.","source":"https://openrouter.ai/terms","confidence":"medium"},"sexual_explicit":{"tos":"conditional","behavior":"conditional","note":"Lawful content only; depends on routed model. Some roleplay models comply. CSAM screened and banned.","source":"https://openrouter.ai/terms","confidence":"medium"},"violence_fiction":{"tos":"conditional","behavior":"allowed","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"profanity":{"tos":"conditional","behavior":"allowed","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"allowed","note":"Provider-dependent; many RP models available.","source":"https://openrouter.ai/terms","confidence":"medium"},"political_opinion":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"gov_criticism":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent; CN-origin models carry their censorship.","source":"https://openrouter.ai/terms","confidence":"medium"},"religion_satire":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"misinfo_generation":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"cbrn":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent; frontier models refuse, uncensored ones may not. Illegal use banned.","source":"https://openrouter.ai/terms","confidence":"medium"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent; red-teaming OpenRouter itself needs written approval.","source":"https://openrouter.ai/terms","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"extremism_analysis":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"Terms bar impersonating others on the service itself.","source":"https://openrouter.ai/terms","confidence":"medium"},"deepfake_realperson":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"pii_doxxing":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"facial_recognition":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"surveillance":{"tos":"conditional","behavior":"conditional","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"copyright_repro":{"tos":"conditional","behavior":"conditional","note":"DMCA process and repeat-infringer policy; model-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"academic_dishonesty":{"tos":"conditional","behavior":"allowed","note":"Provider-dependent.","source":"https://openrouter.ai/terms","confidence":"medium"},"competitor_training":{"tos":"conditional","behavior":"allowed","note":"Must follow each model's terms (e.g. OpenAI/Anthropic restrict competitor training).","source":"https://openrouter.ai/terms","confidence":"medium"},"output_ownership":{"tos":"conditional","behavior":"allowed","note":"Governed by each model's terms.","source":"https://openrouter.ai/terms","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Commercial use allowed subject to model terms.","source":"https://openrouter.ai/terms","confidence":"medium"},"redistribution_maas":{"tos":"restricted","behavior":"allowed","note":"Reselling API access and multi-accounting to dodge limits prohibited.","source":"https://openrouter.ai/terms","confidence":"high"},"finetune_modify":{"tos":"conditional","behavior":"allowed","note":"N/A for the router; model terms apply.","source":"https://openrouter.ai/terms","confidence":"low"},"military_defense":{"tos":"conditional","behavior":"unknown","note":"Model terms apply.","source":"https://openrouter.ai/terms","confidence":"low"},"law_enforcement":{"tos":"conditional","behavior":"unknown","note":"Model terms apply.","source":"https://openrouter.ai/terms","confidence":"low"},"high_stakes_automation":{"tos":"conditional","behavior":"unknown","note":"Model terms apply.","source":"https://openrouter.ai/terms","confidence":"low"},"minimum_age":{"tos":"restricted","behavior":"allowed","note":"Must be at least 18.","source":"https://openrouter.ai/terms","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"unknown","note":"Prompt logging is opt-in. Some providers may train; account setting blocks training providers (free and paid separate). OpenRouter opts out where possible.","source":"https://openrouter.ai/docs/guides/privacy/logging","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"allowed","note":"Restricted Models can't be accessed from certain countries; VPN circumvention is a breach.","source":"https://openrouter.ai/terms","confidence":"high"},"weapons_industry":{"tos":"conditional","behavior":"unknown","note":"Model terms apply.","source":"https://openrouter.ai/terms","confidence":"low"},"autonomous_agents":{"tos":"conditional","behavior":"allowed","note":"Allowed; scraping the site/service and bypassing anti-scraping barred.","source":"https://openrouter.ai/terms","confidence":"medium"},"romantic_companion":{"tos":"conditional","behavior":"allowed","note":"Many roleplay/companion models available; depends on model terms.","source":"https://openrouter.ai/terms","confidence":"medium"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"Red teaming OpenRouter needs written approval; unmoderated models are offered, jailbreaking upstream models may breach their terms.","source":"https://openrouter.ai/terms","confidence":"medium"}}},{"id":"deepseek-service","name":"DeepSeek app / API","vendor":"DeepSeek (Hangzhou DeepSeek AI)","kind":"llmaas","flagship_models":["DeepSeek-V4.1 Flash","DeepSeek-V4 Pro"],"license":"proprietary service","jurisdiction":"CN","policy_docs":[{"title":"Terms of Use","url":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","effective":"2026-03-27"},{"title":"Privacy Policy","url":"https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html","effective":"2026-02-10"}],"default_customizability":"medium","summary":"Hosted DeepSeek is governed by mainland-Chinese law with data stored in the PRC and trains on inputs unless opted out. It hard-censors Tiananmen/Xi/Taiwan topics with an extra output filter on top of model alignment, while the ToS itself is generous on output ownership and distillation.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"restricted","note":"ToS s3.4 bans pornographic content; hosted chat refuses explicit erotica.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"restricted","note":"ToS s3.4 bans pornographic content.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"violence_fiction":{"tos":"restricted","behavior":"allowed","note":"ToS s3.4 bans violent content; fiction is nonetheless generally written.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"profanity":{"tos":"restricted","behavior":"allowed","note":"ToS s3.4(1): no content that is \"offensive, abusive, tortious or vulgar\"; crude language on request is at risk.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"allowed","note":"Not named. s3.4 bans chatbots that promote violence/extremism, instruct crimes or glorify harm; villain personas fine within that.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"Policy silent; only general lawfulness clause applies (s3.2/3.6); PRC law governs the service.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"gov_criticism":{"tos":"conditional","behavior":"prohibited","note":"No national-security clause in ToS; s3.2 bans activities violating laws and PRC law governs, so state-sensitive topics are exposed to legal limits.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"religion_satire":{"tos":"conditional","behavior":"conditional","note":"s3.4(4) bans content discriminating on religion; s3.4(1) bans offensive content. Satire not named.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"hate_in_context":{"tos":"prohibited","behavior":"conditional","note":"ToS s3.4 bans hateful/discriminatory content; analysis contexts answered.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"misinfo_generation":{"tos":"restricted","behavior":"conditional","note":"ToS s3.4 bans defamatory content; no explicit disinformation clause found.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Policy silent; only general lawfulness clause applies; s3.1 requires labelling published outputs as AI-generated.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Not banned; s5 says Outputs are not professional medical advice and users must consult professionals.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Not banned; s5 says Outputs are not professional legal advice and users must consult professionals.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"Not banned; s5 says Outputs are not professional financial advice and users must consult professionals.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Policy silent on self-harm; s3.4(3) only bans content that harasses or harms another. General lawfulness applies.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"No drug clause; s3.4(9) bans content/instructions constituting a criminal offence (general illegality clause).","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"restricted","note":"No weapons clause; s3.4(9) bans instructions for a criminal offence (general illegality clause).","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"cbrn":{"tos":"restricted","behavior":"prohibited","note":"s3.6(2) bans \"dangerous purposes that may have serious harmful impacts on physical health... society\"; s3.4(6) bans terrorism content.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"offensive_security":{"tos":"restricted","behavior":"conditional","note":"s3.5 bans generating/spreading malicious programs and unauthorised intrusion/vuln scanning; legitimate pentest help not named.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"No clause on lockpicking/shoplifting; s3.4(9) bans criminal-offence instructions (general illegality clause).","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"s3.4(6) bans content that facilitates, promotes or glorifies terrorist/extremism content; neutral analysis not addressed.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"impersonation":{"tos":"restricted","behavior":"conditional","note":"ToS s3.4 bans unlabeled impersonation; labelled personas allowed.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"deepfake_realperson":{"tos":"restricted","behavior":"restricted","note":"s3.4(10) bans impersonating a real person unless labelled parody/unofficial or with consent; s3.4(5) bans sexual content.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"pii_doxxing":{"tos":"conditional","behavior":"restricted","note":"No doxxing clause; s3.2 bars infringing others' legal rights; s1.6 urges not sharing others' personal data without consent.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no biometric clause.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"surveillance":{"tos":"restricted","behavior":"conditional","note":"s3.6 bans illegal uses; PRC law applies; no surveillance-specific clause found.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"copyright_repro":{"tos":"conditional","behavior":"conditional","note":"s3.6(3) bans activities infringing IP rights; reproduction of copyrighted text not otherwise addressed.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Policy silent; only general lawfulness clause applies; no academic-integrity clause.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"ToS 4.2(3) expressly permits training other models (distillation) if lawful; s3.5 bans reverse engineering, 3.6 unfair competition.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"ToS 4.2: DeepSeek assigns any rights in Outputs to user; commercial use fine.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"high"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Service and API usable commercially; PRC mainland law governs (s10.1), Hangzhou courts.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"high"},"redistribution_maas":{"tos":"restricted","behavior":"allowed","note":"Reselling the service needs agreement; API reselling widespread.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Not applicable to service; API has no fine-tuning of weights.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"military_defense":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no military clause in ToS.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no policing clause.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"high_stakes_automation":{"tos":"conditional","behavior":"unknown","note":"s5: Outputs used for credit, education, employment, housing, insurance, legal or medical decisions about people must get human review.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"minimum_age":{"tos":"conditional","behavior":"allowed","note":"Intended for adults; under-18 only with guardian consent (s2.1). Privacy policy: 14+ in EEA/UK.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Inputs/outputs used to improve service (de-identified); opt-out via 'Improve the model for everyone'. Data stored in PRC. API: same terms.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-privacy-policy.html","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"No ToS exclusion, but banned on gov devices/app stores in Italy, South Korea (paused), others; data in PRC.","source":"https://en.wikipedia.org/wiki/DeepSeek","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no weapons-industry clause.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agentic/tool use is a headline feature of current models.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"medium"},"romantic_companion":{"tos":"restricted","behavior":"conditional","note":"ToS bans pornographic content and unlabeled impersonation; companion personas not named. Hosted app plays friendly personas, sexual escalation refused.","source":"https://cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html","confidence":"low"},"safety_bypass":{"tos":"prohibited","behavior":"restricted","note":"ToS s3.5 bans reverse engineering/circumventing; hosted service has extra output filter that deletes sensitive answers, so state-topic refusals are hard to override.","source":"https://arxiv.org/abs/2505.12625","confidence":"low"}}},{"id":"qwen-service","name":"Qwen Chat / Alibaba Model Studio","vendor":"Alibaba (Qwen)","kind":"llmaas","flagship_models":["Qwen3.8-Max","Qwen3.6-Plus"],"license":"proprietary service","jurisdiction":"CN","policy_docs":[{"title":"Qwen Terms of Service","url":"https://qwen.ai/termsservice","effective":"unknown"},{"title":"Alibaba Cloud Model Studio product terms","url":"https://www.alibabacloud.com/help/en/legal/latest/alibaba-cloud-international-website-product-terms-of-service-model-studio","effective":"unknown"},{"title":"Policy read for gap-fill","url":"https://qwen.ai/usagepolicy","effective":"2026-10-20 (fetched)"}],"default_customizability":"medium","summary":"Hosted Qwen applies PRC-aligned filtering on top of the model, with extra API-level blocking observed on China-critical prompts. Primary ToS pages were JS-rendered or 404 for the fetch tool, so most ToS cells are unknown.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"restricted","note":"Use Policy III.2(1): no \"obscene, pornographic, graphic\" content; adult erotica not carved out.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"prohibited","note":"Use Policy III.2(1) bans disseminating or depicting \"obscene, pornographic, graphic\" content.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"violence_fiction":{"tos":"restricted","behavior":"allowed","note":"Use Policy III.2(1) bans depicting \"graphic... terrorist, or violent extremist acts\"; fictional violence not addressed.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Use Policy only \"encourages\" civilized language and avoiding vulgar remarks (I.2); not a red line, except hateful content.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"allowed","note":"Not named; red lines III.2-4 ban crime, violence, hatred and self-harm promotion, so dark characters allowed within them.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"political_opinion":{"tos":"conditional","behavior":"restricted","note":"No opinion clause; III.1 bans content endangering national security and social stability, so contested politics is exposed.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"gov_criticism":{"tos":"restricted","behavior":"prohibited","note":"Use Policy III.1: no content that \"endangers national security\", \"undermines the national system\" or similar; \"social stability\".","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"religion_satire":{"tos":"conditional","behavior":"conditional","note":"III.4(1) bans content that disparages or stereotypes people by \"religious beliefs\"; satire of ideas not named.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"hate_in_context":{"tos":"prohibited","behavior":"conditional","note":"III.4: no content that disparages, stereotypes, demeans or incites hostility toward protected groups.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"conditional","note":"III.5(2): \"Do not generate or publish any misleading, fraudulent, or false information\"; AI-labelling encouraged.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"election_campaign":{"tos":"restricted","behavior":"conditional","note":"III.1(1) bans content that \"undermines the fairness of local elections\"; campaign material otherwise not named.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"ToS II.2 bans giving professional advice (e.g. doctors) without qualifications; V.3 says output is no substitute for professionals.","source":"https://qwen.ai/termsservice","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"ToS II.2 bans giving professional advice (e.g. lawyers) without qualifications; V.3 says output is no substitute for professionals.","source":"https://qwen.ai/termsservice","confidence":"medium"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"Not banned; Use Policy I.4 tells users to verify finance/investment output; ToS V.3 says no substitute for professional advice.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"III.4(3) bans content that encourages or instigates self-harm/suicide; supportive discussion not addressed.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"III.2(2) bans drug trafficking and conduct prohibited by local law (general illegality clause); drug information not addressed.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"restricted","note":"III.2(2) bans weapons trafficking and conduct prohibited by local law (general illegality clause).","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"cbrn":{"tos":"restricted","behavior":"prohibited","note":"No CBRN clause; III.1(2) bans content endangering public safety and III.2(1) bans terrorist acts.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"offensive_security":{"tos":"conditional","behavior":"conditional","note":"ToS II.2 bans probing the Qwen service; generating malware not named, only \"illegal, harmful, or abusive activities\".","source":"https://qwen.ai/termsservice","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"No lockpicking/shoplifting clause; III.2(2) bans conduct prohibited by local law (general illegality clause).","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"III.2(1) and III.4 ban disseminating terrorist/extremist acts or hateful content; critical analysis not addressed.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"ToS II.2 bans falsely presenting AI output as human; Use Policy III.6 bans identity impersonation to attack the service.","source":"https://qwen.ai/termsservice","confidence":"medium"},"deepfake_realperson":{"tos":"restricted","behavior":"restricted","note":"Use Policy III.3(3) bans AI-generated or manipulated intimate images; ToS bans violating portrait/reputation rights.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"pii_doxxing":{"tos":"restricted","behavior":"restricted","note":"ToS II.2 bans infringing privacy rights; Use Policy III.2(2) lists \"violations of citizens' privacy\".","source":"https://qwen.ai/termsservice","confidence":"medium"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; only portrait-rights infringement is barred.","source":"https://qwen.ai/termsservice","confidence":"medium"},"surveillance":{"tos":"conditional","behavior":"conditional","note":"No surveillance clause; privacy violations barred (ToS II.2, Use Policy III.2(2)).","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"copyright_repro":{"tos":"conditional","behavior":"conditional","note":"ToS II.2 bans infringing copyright/IP; Use Policy III.5(1) bans infringing content for profit.","source":"https://qwen.ai/termsservice","confidence":"medium"},"academic_dishonesty":{"tos":"prohibited","behavior":"allowed","note":"Use Policy III.5(3): no plagiarism or academic misconduct, \"ghostwriting, cheating on exams, data fabrication\".","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"competitor_training":{"tos":"prohibited","behavior":"unknown","note":"ToS II.2: no using the Service or its content to develop or train competing AI/ML models.","source":"https://qwen.ai/termsservice","confidence":"medium"},"output_ownership":{"tos":"conditional","behavior":"unknown","note":"ToS IV.3: Qwen assigns any rights in Output to the user, but Service is \"private and non-commercial use only\" (I.4). Model Studio terms not read.","source":"https://qwen.ai/termsservice","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Model Studio API is commercial paid service; Alibaba Cloud International terms (Singapore entity) apply outside mainland.","source":"https://www.alibabacloud.com/help/en/legal/latest/alibaba-cloud-international-website-product-terms-of-service-model-studio","confidence":"low"},"redistribution_maas":{"tos":"restricted","behavior":"unknown","note":"Consumer ToS II.2 bans modifying, leasing, selling, distributing the Service and bot access. Model Studio API terms not readable.","source":"https://qwen.ai/termsservice","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"unknown","note":"Hosted service only; ToS II.2 bans modifying the Service. Weights fine-tuning falls under separate open-weights licences.","source":"https://qwen.ai/termsservice","confidence":"medium"},"military_defense":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; ToS only bars use in sanctioned regions/export-controlled ends.","source":"https://qwen.ai/termsservice","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no policing clause.","source":"https://qwen.ai/termsservice","confidence":"medium"},"high_stakes_automation":{"tos":"conditional","behavior":"unknown","note":"ToS II.2: output on individuals must not be used for credit, education, employment, housing, insurance, legal or healthcare decisions without human review.","source":"https://qwen.ai/termsservice","confidence":"medium"},"minimum_age":{"tos":"conditional","behavior":"unknown","note":"ToS I.1: 18 or the minimum age to consent in your location, whichever is greater.","source":"https://qwen.ai/termsservice","confidence":"medium"},"trains_on_your_data":{"tos":"conditional","behavior":"unknown","note":"ToS IV.5: content may be used to train models; opt-out by emailing privacy@qwen.com. Model Studio API policy not read.","source":"https://qwen.ai/termsservice","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Alibaba Cloud sanctions/export compliance; separate China and International sites with different terms. Not fully verified.","source":"https://www.alibabacloud.com/help/en/legal/latest/alibaba-cloud-international-website-product-terms-of-service-model-studio","confidence":"low"},"weapons_industry":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; no weapons-industry clause.","source":"https://qwen.ai/termsservice","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agentic/tool use is a headline feature of current models.","source":"https://www.alibabacloud.com/help/en/legal/latest/alibaba-cloud-international-website-product-terms-of-service-model-studio","confidence":"medium"},"romantic_companion":{"tos":"restricted","behavior":"conditional","note":"Pornographic content banned (Use Policy III.2); companion personas not named. Romance itself is not barred.","source":"https://qwen.ai/usagepolicy","confidence":"medium"},"safety_bypass":{"tos":"prohibited","behavior":"restricted","note":"Use Policy III.6 bans jailbreak attacks, evading detection with variants/homophones and manipulating models past safety mechanisms.","source":"https://qwen.ai/usagepolicy","confidence":"medium"}}},{"id":"kimi-service","name":"Kimi app / Moonshot API","vendor":"Moonshot AI","kind":"llmaas","flagship_models":["Kimi K3","Kimi K2.6"],"license":"proprietary service","jurisdiction":"CN (platform terms under Singapore law)","policy_docs":[{"title":"Platform Terms (modeluse)","url":"https://platform.kimi.ai/docs/agreement/modeluse","effective":"2026-07-30"},{"title":"Kimi user privacy agreement (CN app)","url":"https://www.kimi.com/user/agreement/userPrivacy","effective":"2024-12-19"},{"title":"Policy read for gap-fill","url":"https://www.kimi.com/user/agreement/modelUse","effective":"2026-10-20 (fetched)"}],"default_customizability":"medium","summary":"The international API platform uses an EU-AI-Act-flavoured prohibited-use list under Singapore law, bans competing-model development and trains on customer content by default. The consumer app follows PRC-aligned filtering on sensitive topics.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"conditional","note":"Consumer agreement bans \"obscene, pornographic\" content (淫秽、色情); API s3.1(5) bans sexualising children. Adult fiction not carved out.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"sexual_explicit":{"tos":"restricted","behavior":"restricted","note":"Platform terms ban unlawful harmful content; sexual content refused in app.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"low"},"violence_fiction":{"tos":"conditional","behavior":"allowed","note":"Consumer agreement bans promoting violence/murder (暴力、凶杀); API s3.1(5) bans promoting violence. Fiction not addressed.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"profanity":{"tos":"conditional","behavior":"allowed","note":"Consumer agreement bans publicly insulting or defaming others; API s3.1(5) bans bullying/harassment. Crude language otherwise unaddressed.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"allowed","note":"Not named; bans on inciting crime, violence, hatred and terrorism apply (API s3.1, consumer terms).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"political_opinion":{"tos":"conditional","behavior":"restricted","note":"No opinion clause; consumer agreement bans inciting subversion or damaging state-organ credibility, so contested politics is exposed.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"gov_criticism":{"tos":"restricted","behavior":"restricted","note":"Consumer: no incitement to \"subvert state power\", \"split the country\" or \"damage the credibility of state organs\" (损害国家机关信誉).","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"religion_satire":{"tos":"conditional","behavior":"conditional","note":"Consumer bans discrimination based on race, gender or religion; API bans hatred. Satire of ideas not named.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"hate_in_context":{"tos":"prohibited","behavior":"conditional","note":"Consumer bans ethnic hatred/discrimination and race, gender, religion bias; API s3.1(5) bans promoting hatred.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"misinfo_generation":{"tos":"restricted","behavior":"conditional","note":"Removing AI-content labels banned; deceptive manipulation techniques banned.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"API s3.4(1) bans manipulative techniques that distort behaviour; campaign material otherwise not named.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"HIPAA PHI processing prohibited; general info given.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"Consumer bars use in services needing qualification, incl. legal services; API: output not a substitute for professional legal advice.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"API s4: output not a substitute for professional financial advice; not banned. Consumer bars qualification-required services.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"No self-harm clause; API s3.1(1) bans uses harmful to physical or spiritual health, s3.1(5) promoting others' suffering.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"API s3.1(2) bans developing/distributing illegal substances; drug information otherwise falls under the general illegality clause.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"restricted","note":"No weapons clause; API s3.1(7) and consumer ban uses prohibited by law (general illegality clause).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"cbrn":{"tos":"restricted","behavior":"prohibited","note":"No CBRN clause; API s3.1(1)-(2) bans uses harming physical health and terrorism; consumer bans terror/violence content.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"offensive_security":{"tos":"restricted","behavior":"conditional","note":"API s3.2(1)-(2) bans unauthorised network access and providing tools designed for intrusion or data theft; malware named in s3.3(2).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"No lockpicking/shoplifting clause; API s3.1(7) bans uses prohibited by law (general illegality clause).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"API s3.1(2) bans promoting terrorism; consumer bans terror content. Neutral analysis not addressed.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"API s3.1(4)/3.4(1) ban fraudulent or deceptive use and manipulation; labelled personas not specifically addressed.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"deepfake_realperson":{"tos":"restricted","behavior":"restricted","note":"API s3.1(4),(6) ban deception and compromising others' privacy; s3.3(3) bans removing AI-content labels.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"pii_doxxing":{"tos":"restricted","behavior":"restricted","note":"API s3.1(6): no \"compromising the privacy of others\"; personal data needs notices and consent.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"facial_recognition":{"tos":"restricted","behavior":"unknown","note":"s3: real-time remote biometric ID for law enforcement and non-consensual face databases banned; emotion inference in work/school restricted.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"surveillance":{"tos":"restricted","behavior":"unknown","note":"Social scoring, biometrics-based surveillance banned.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"copyright_repro":{"tos":"conditional","behavior":"conditional","note":"API s3.1(3) bans IP-infringing activity; reproducing copyrighted text not otherwise addressed.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Policy silent; only general lawfulness clause applies (API and consumer terms).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"competitor_training":{"tos":"prohibited","behavior":"unknown","note":"API terms s3.2(5): developing or serving competing products/models without authorization prohibited; s3.2(4) no reverse engineering.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"high"},"output_ownership":{"tos":"conditional","behavior":"unknown","note":"API s4: Moonshot claims no ownership of content. Consumer agreement allows only non-commercial personal use; commercial use of outputs needs written consent.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"API commercial use allowed; Singapore law, SIAC arbitration (s12) for platform terms (July 30, 2026).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"high"},"redistribution_maas":{"tos":"prohibited","behavior":"allowed","note":"s3.2: reselling/sublicensing service without authorization, trading API keys prohibited.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"high"},"finetune_modify":{"tos":"conditional","behavior":"conditional","note":"Service-side only; no weights access.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"low"},"military_defense":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies (API and consumer terms).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"law_enforcement":{"tos":"restricted","behavior":"unknown","note":"Real-time biometric ID for law enforcement and criminal-risk prediction banned.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"high_stakes_automation":{"tos":"restricted","behavior":"unknown","note":"s3: social scoring and predicting criminal risk from personal traits banned.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"minimum_age":{"tos":"conditional","behavior":"allowed","note":"Platform: 18 or local minimum age; minors need guardian consent; under 14 no personal info (s2, 3.4).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"high"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"s4: Customer Content may be used to improve services unless written agreement/enterprise arrangement says otherwise. Consumer: de-identified chats used.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Comprehensive-sanction countries and listed persons excluded (s2, 13).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"high"},"weapons_industry":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies (API and consumer terms).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"autonomous_agents":{"tos":"conditional","behavior":"allowed","note":"Agentic use supported (Kimi agent mode); data extraction outside APIs prohibited.","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"},"romantic_companion":{"tos":"restricted","behavior":"conditional","note":"Pornographic content banned (consumer); companion personas not named in API or consumer terms.","source":"https://www.kimi.com/user/agreement/modelUse","confidence":"medium"},"safety_bypass":{"tos":"prohibited","behavior":"conditional","note":"Platform terms s3 ban evading safety detection and removing AI-content labels; no official relaxation process (enterprise agreement only for data terms).","source":"https://platform.kimi.ai/docs/agreement/modeluse","confidence":"medium"}}},{"id":"glm-service","name":"Z.ai chat / GLM API","vendor":"Zhipu AI (Z.ai)","kind":"llmaas","flagship_models":["GLM-5.3","GLM-5.2"],"license":"proprietary service","jurisdiction":"CN (terms under Singapore law)","policy_docs":[{"title":"Z.ai Terms of Service","url":"https://chat.z.ai/legal-agreement/terms-of-service","effective":"2026-04-14"}],"default_customizability":"medium","summary":"Z.ai's international terms are Singapore-law with SIAC arbitration, but retain a broad prohibited list including national-security/'any country's interests' content, political campaign content, and regulated-decision use. The consumer app does not train on personal data but API data is not used for training without consent.","cells":{"sexual_fiction":{"tos":"restricted","behavior":"restricted","note":"ToS bans obscene/sexually explicit material.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"sexual_explicit":{"tos":"prohibited","behavior":"restricted","note":"ToS bans pornographic or sexually explicit material.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"high"},"violence_fiction":{"tos":"restricted","behavior":"allowed","note":"ToS bans violent material; fiction mostly written.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"low"},"profanity":{"tos":"conditional","behavior":"allowed","note":"III.3(h) bans abusive, insulting, hateful conduct toward third parties; crude language on request not named.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"dark_roleplay":{"tos":"conditional","behavior":"allowed","note":"Not named; III.3(f),(c),(k) ban violent, terroristic, hateful and self-harm content. III.6(d): virtual characters must not infringe rights.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"political_opinion":{"tos":"restricted","behavior":"restricted","note":"ToS bans political campaign content; model neutral/deflecting.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"gov_criticism":{"tos":"restricted","behavior":"restricted","note":"ToS bans content threatening national security or harming 'any country's interests'; model also aligned on PRC topics.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"religion_satire":{"tos":"conditional","behavior":"conditional","note":"III.3(d) bans \"undermining freedom of religion or inciting discrimination based on religious beliefs\"; satire not named.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"hate_in_context":{"tos":"prohibited","behavior":"conditional","note":"ToS bans hate/discrimination/extremism.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"conditional","note":"ToS bans false information that could mislead the public.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"high"},"election_campaign":{"tos":"prohibited","behavior":"restricted","note":"ToS prohibits political campaign content.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"high"},"medical_advice":{"tos":"restricted","behavior":"conditional","note":"ToS bars use for regulated medical determinations.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"legal_advice":{"tos":"restricted","behavior":"conditional","note":"ToS bars use for regulated legal determinations.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"financial_advice":{"tos":"restricted","behavior":"conditional","note":"ToS bars use for regulated financial determinations.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"self_harm_support":{"tos":"restricted","behavior":"conditional","note":"ToS bans content promoting self-harm.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"conditional","note":"No drug clause; III.3(m) bans acts prohibited by applicable law (general illegality clause).","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"restricted","note":"No weapons clause; III.3(m) bans acts prohibited by applicable law (general illegality clause).","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"cbrn":{"tos":"restricted","behavior":"prohibited","note":"No CBRN clause; III.3(a),(c) bans threatening critical infrastructure/national security and promoting terrorism.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"offensive_security":{"tos":"prohibited","behavior":"conditional","note":"API terms 1.f.xiii bar generating malicious code; III.4(a) bans intentionally spreading malicious programs.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"No lockpicking/shoplifting clause; III.3(m) bans acts prohibited by law (general illegality clause).","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"III.3(c) bans \"inciting, facilitating, or promoting violent extremism, terrorism\"; analysis not addressed.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"III.5(a) bans passing off AI output as human; III.6(d) virtual characters must not infringe third-party rights.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"deepfake_realperson":{"tos":"restricted","behavior":"restricted","note":"III.3(e),(i),(g) ban misleading false information, unauthorised personal-data generation and sexually explicit content.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"ToS bans unauthorized collection/exposure of others' personal data.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"high"},"facial_recognition":{"tos":"restricted","behavior":"unknown","note":"III.3(i) bans collecting, inferring or generating others' personal identification information without authorisation.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"surveillance":{"tos":"conditional","behavior":"conditional","note":"No surveillance clause; III.3(i) bans unauthorised collection or inference of others' personal data; III.4(b) bans monitoring via bots.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"copyright_repro":{"tos":"conditional","behavior":"conditional","note":"III.3(h) and V bar infringing third-party IP; reproducing copyrighted text otherwise unaddressed.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Policy silent; only general lawfulness clause applies; III.6(b) only restricts replacing education professionals.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"competitor_training":{"tos":"prohibited","behavior":"unknown","note":"ToS III.4.f and API terms 1.f.xii bar using Z.ai (models, prompts, outputs) to build competing models.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"high"},"output_ownership":{"tos":"allowed","behavior":"unknown","note":"IV.4: user retains all rights in Prompts and Outputs generated at their request; Z.ai asserts no ownership.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"API commercial use allowed; Singapore law and SIAC arbitration (XIII.3).","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"redistribution_maas":{"tos":"restricted","behavior":"unknown","note":"III/V: no lease, copying or sublicensing without written consent; resale to restricted persons/countries banned.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"finetune_modify":{"tos":"conditional","behavior":"unknown","note":"API terms allow fine-tuning on the platform; 1.f.xii bans using outputs to train or fine-tune external/competing models.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"military_defense":{"tos":"prohibited","behavior":"unknown","note":"X.7: service \"shall not be used for any prohibited end use, including military purposes or end uses involving human rights violations\".","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"unknown","note":"Policy silent; only general lawfulness clause applies; III.6(a) only bars high-risk automated decisions.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"high_stakes_automation":{"tos":"restricted","behavior":"unknown","note":"ToS bars use for regulated medical/legal/financial/educational determinations.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"minimum_age":{"tos":"conditional","behavior":"allowed","note":"No set minimum; minors should use with guardian (IX.8).","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"trains_on_your_data":{"tos":"conditional","behavior":"conditional","note":"Consumer: non-personal User Content may improve models (IV.3.a). API: not used unless you opt in (IV.7).","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"high"},"geo_restrictions":{"tos":"conditional","behavior":"unknown","note":"Content processed outside user's jurisdiction; sanctions compliance not verified.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"low"},"weapons_industry":{"tos":"restricted","behavior":"unknown","note":"X.7 bans military end use; weapons industry not named explicitly.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"medium"},"autonomous_agents":{"tos":"conditional","behavior":"allowed","note":"Agentic features supported; automation of other services not specifically verified.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"low"},"romantic_companion":{"tos":"restricted","behavior":"conditional","note":"ToS bans sexually explicit material and content inappropriate for minors; sustained romantic personas not named.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"low"},"safety_bypass":{"tos":"prohibited","behavior":"conditional","note":"Z.ai terms ban abusing/circumventing service controls (not verified verbatim); no official relaxation process found.","source":"https://chat.z.ai/legal-agreement/terms-of-service","confidence":"low"}}},{"id":"openai-gpt-oss","name":"gpt-oss (OpenAI open weights)","vendor":"OpenAI","kind":"open-weights","flagship_models":["gpt-oss-120b","gpt-oss-20b"],"license":"Apache-2.0 + gpt-oss usage policy (comply with law)","jurisdiction":"US","policy_docs":[{"title":"gpt-oss-120b model card on Hugging Face","url":"https://huggingface.co/openai/gpt-oss-120b","effective":"2025-08-05"},{"title":"gpt-oss usage policy","url":"https://huggingface.co/openai/gpt-oss-120b/raw/main/USAGE_POLICY","effective":"2025-08-05"},{"title":"gpt-oss model card (arXiv 2508.10925)","url":"https://arxiv.org/abs/2508.10925","effective":"2025-08-05"}],"default_customizability":"high","summary":"Permissive Apache-2.0 weights whose usage policy only asks users to follow applicable law, so legally anything goes; the instruct models ship with OpenAI-style safety training that refuses CBRN, explicit sexual content and malware by default but can be fine-tuned away.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"Apache-2.0 plus a policy that only demands legal compliance; default checkpoint refuses much explicit content but writes mild romance.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"No licence ban; chat model refuses graphic sexual content by default, abliterated/fine-tuned variants common.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"No restriction; model writes fictional violence.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No restriction; complies with crude language on request.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"conditional","note":"No restriction; model often hedges or refuses on extreme dark scenarios but generally plays villains.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"No restriction; trained to stay neutral and avoid taking sides.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction; answers on Tiananmen etc. though reports of occasional refusals/hallucination in SpeechMap-type tests.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"No restriction; typically complies with mild religion jokes, refuses hateful ones.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"hate_in_context":{"tos":"allowed","behavior":"conditional","note":"No restriction; classification/analysis fine, hateful generation refused.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"misinfo_generation":{"tos":"allowed","behavior":"conditional","note":"Not restricted by licence beyond 'comply with law'; model sometimes writes labelled fiction, refuses operational disinformation.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"No restriction; model may help with generic persuasive text.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"allowed","note":"No restriction; model card warns it is not for medical diagnosis (HealthBench evaluated).","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"allowed","note":"No restriction; answers with caveats.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"allowed","note":"No restriction; answers with caveats.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"No restriction; trained on OpenAI safe-completion policy: supportive, no method detail.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"drugs":{"tos":"allowed","behavior":"conditional","note":"No restriction; harm-reduction info given, synthesis refused.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"No licence restriction; instruct model refuses build instructions per OpenAI safety training.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"cbrn":{"tos":"allowed","behavior":"prohibited","note":"No licence ban, but model trained to refuse; OpenAI ran worst-case malicious fine-tuning study (Preparedness) finding no High capability reached.","source":"https://arxiv.org/abs/2508.10925","confidence":"high"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"No restriction; refuses clear malware requests, helps with defensive/educational security.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"No restriction; answers hobby lockpicking, refuses crime-intent.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"No restriction; analysis allowed.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"No restriction; persona roleplay works, hedges on real-person deceptive use.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"deepfake_realperson":{"tos":"allowed","behavior":"restricted","note":"No restriction in licence (text only); model refuses sexual content about real people.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"No restriction; no web access so little capability; refuses doxxing.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"facial_recognition":{"tos":"allowed","behavior":"allowed","note":"Text-only model; no vision, so not applicable.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No restriction; model may refuse stalking assistance.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"copyright_repro":{"tos":"allowed","behavior":"restricted","note":"No restriction; model trained to refuse lyrics/long copyrighted text, though it memorises less than peers.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No restriction; writes essays.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 contains no anti-distillation clause; outputs may train other models.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 places no claim on outputs.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 plus patent grant, no user-count thresholds.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"redistribution_maas":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 permits redistribution and hosting, keep license/NOTICE.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Fine-tuning and modification explicitly permitted; OpenAI states it expects adversarial fine-tuning and tested it.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"No field-of-use limits in Apache-2.0; usage policy only requires compliance with law.","source":"https://huggingface.co/openai/gpt-oss-120b/raw/main/USAGE_POLICY","confidence":"high"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"No field-of-use limits.","source":"https://huggingface.co/openai/gpt-oss-120b/raw/main/USAGE_POLICY","confidence":"high"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"No licence limits; model card advises against unsupervised high-stakes use.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age gate for weights; downloader platforms (Hugging Face 13+) apply their own terms.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted weights send nothing to OpenAI; third-party hosts have their own policies.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"No geographic restriction in Apache-2.0; export laws still apply.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No field-of-use limit; law compliance only.","source":"https://huggingface.co/openai/gpt-oss-120b/raw/main/USAGE_POLICY","confidence":"high"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Built for agentic tool use (browsing, Python, function calling); no restriction.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"high"},"romantic_companion":{"tos":"allowed","behavior":"conditional","note":"No licence restriction; default chat model plays mild romantic roleplay but refuses explicit content.","source":"https://huggingface.co/openai/gpt-oss-120b","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Fine-tuning/abliteration permitted by Apache-2.0; OpenAI explicitly tested adversarial fine-tuning. Default model moderately jailbreakable; refusals easily removed by fine-tuning.","source":"https://arxiv.org/abs/2508.10925","confidence":"medium"}}},{"id":"meta-llama","name":"Llama (Meta open weights)","vendor":"Meta","kind":"open-weights","flagship_models":["Llama 4 Maverick","Llama 4 Scout","Muse Glimmer 30B (Apache-2.0, reported)"],"license":"Custom (Llama 4 Community License + AUP)","jurisdiction":"US","policy_docs":[{"title":"Llama 4 Community License","url":"https://dev.meta.ai/llama/llama4/license/","effective":"2025-04-05"},{"title":"Llama 4 Acceptable Use Policy","url":"https://dev.meta.ai/llama/llama4/use-policy/","effective":"2025-04-05 (page as fetched)"}],"default_customizability":"high","summary":"Weights usable commercially free below 700M MAU, with 'Built with Llama' attribution and 'Llama' naming for derivatives. AUP bans military/weapons (government carve-out added Nov 2024) and multimodal rights are withheld from EU-domiciled entities. Newest Meta frontier models (Muse Spark) are closed; Muse Glimmer 30B reportedly Apache-2.0 (Aug 2026, unverified).","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"Licence silent; instruct models refuse more than community finetunes but write romance.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"AUP bans sexual solicitation/CSAM only; Llama 4 chat usually declines explicit text; easily finetuned.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Complies in fiction.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Llama 4 pitched as less preachy; complies mostly.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"Meta says Llama 4 tuned to be balanced and refuse less on contested topics.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No limits in licence; behaviour mostly open.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"AUP bans harassment/discrimination; analysis contexts fine.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"conditional","note":"AUP bans fraud/disinformation and presenting outputs as human; model writes labelled fiction.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"conditional","note":"Disinformation banned; legitimate campaign copy not specifically restricted.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"low"},"medical_advice":{"tos":"restricted","behavior":"conditional","note":"AUP bans unauthorized practice of medicine; model gives info with disclaimers.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"legal_advice":{"tos":"restricted","behavior":"conditional","note":"AUP bans unauthorized practice of law.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"financial_advice":{"tos":"restricted","behavior":"conditional","note":"AUP bans unauthorized financial practice.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"self_harm_support":{"tos":"restricted","behavior":"conditional","note":"AUP bans enabling self-harm; model offers resources.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"drugs":{"tos":"restricted","behavior":"conditional","note":"AUP bans illegal drugs facilitation; harm-reduction answered.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"weapons_conventional":{"tos":"prohibited","behavior":"restricted","note":"AUP bans guns/illegal weapons and weapon development.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"prohibited","note":"AUP bans nuclear, chemical, biological weapons uses.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"high"},"offensive_security":{"tos":"prohibited","behavior":"conditional","note":"AUP bans malware creation; model helps with defensive security.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"conditional","note":"Not specifically addressed; illegal activity clause applies.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis allowed; AUP bans terrorism content.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"impersonation":{"tos":"prohibited","behavior":"conditional","note":"AUP bans impersonation without consent.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"Prohibited by AUP; local users can bypass.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"AUP bans collecting/inferring private info without legal right.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"facial_recognition":{"tos":"restricted","behavior":"restricted","note":"Covered by private-info clause; models decline identifying people.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"low"},"surveillance":{"tos":"prohibited","behavior":"conditional","note":"Espionage banned; clauses on private info.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"low"},"copyright_repro":{"tos":"prohibited","behavior":"conditional","note":"AUP bans infringement; model usually declines long lyrics.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No restriction.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"competitor_training":{"tos":"conditional","behavior":"conditional","note":"Llama 4 licence has no ban on training competitors, but derived models must be named 'Llama...'. (Llama 2 banned use for other LLMs.)","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"You own derivatives/outputs; Meta keeps Llama materials.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"high"},"commercial_use":{"tos":"conditional","behavior":"conditional","note":"Free commercial use unless >700M MAU in prior month; then Meta licence at sole discretion.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"high"},"redistribution_maas":{"tos":"conditional","behavior":"conditional","note":"Allowed with licence copy, 'Built with Llama' and Notice file; hosting allowed under same conditions.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Allowed; derivative must keep licence and Llama naming prefix.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"high"},"military_defense":{"tos":"conditional","behavior":"conditional","note":"AUP still lists military/warfare; Nov 2024 Meta made Llama available to US government, defence contractors and allied (Five Eyes etc.) agencies.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"law_enforcement":{"tos":"conditional","behavior":"conditional","note":"No dedicated ban; discrimination in essential services barred.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"low"},"high_stakes_automation":{"tos":"conditional","behavior":"conditional","note":"AUP bans discrimination in employment, credit, housing; failing to disclose known dangers.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age clause for downloaders; consumer products are separate.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"low"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Local weights: no telemetry; N/A for self-hosted.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"conditional","note":"Llama 4: multimodal rights not granted to EU-domiciled individuals/companies; end users of products OK. Export-control compliance.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"high"},"weapons_industry":{"tos":"prohibited","behavior":"prohibited","note":"AUP bans weapon development and ITAR-controlled uses (except government carve-out).","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"No restriction; critical-infrastructure operation banned.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"},"romantic_companion":{"tos":"allowed","behavior":"conditional","note":"No licence ban on companion apps; instruct model plays along mildly, community finetunes fully. Meta itself restricts teen access in its own products.","source":"https://dev.meta.ai/llama/llama4/license/","confidence":"medium"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"Licence/AUP bars circumventing safety measures in products, but weights can be finetuned or abliterated; Llama Guard/Prompt Guard are optional.","source":"https://dev.meta.ai/llama/llama4/use-policy/","confidence":"medium"}}},{"id":"google-gemma","name":"Gemma (Google open weights)","vendor":"Google DeepMind","kind":"open-weights","flagship_models":["Gemma 4 31B (Apache-2.0)","Gemma 3 (custom terms)"],"license":"Gemma 4: Apache-2.0 (HF license field); Gemma 1-3: Custom Gemma Terms of Use + Prohibited Use Policy","jurisdiction":"US","policy_docs":[{"title":"Gemma Terms of Use (Gemma 1-3)","url":"https://ai.google.dev/gemma/terms","effective":"2026-04-01"},{"title":"Gemma Prohibited Use Policy","url":"https://ai.google.dev/gemma/prohibited_use_policy","effective":"2024-02-21"},{"title":"Gemma 4 licence (Apache 2.0)","url":"https://ai.google.dev/gemma/apache_2","effective":"2026-04-01"},{"title":"gemma-4-31B-it model card","url":"https://huggingface.co/google/gemma-4-31b-it","effective":"2026"}],"default_customizability":"high","summary":"Confirmed: Gemma 4 ships under Apache-2.0 per HF metadata, and the Gemma Terms page says Gemma 4 is excluded from its custom terms. Earlier Gemma versions keep the custom terms with a flow-down Prohibited Use Policy. Instruct checkpoint is safety-tuned but easily modified.","cells":{"sexual_fiction":{"tos":"conditional","behavior":"conditional","note":"Gemma 1-3 PUP bans sexually explicit content incl. sexual chatbots (art exceptions). Gemma 4 (Apache-2.0) carries no binding PUP in licence text. Instruct models fairly tame.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"sexual_explicit":{"tos":"restricted","behavior":"restricted","note":"PUP (Gemma 1-3) prohibits pornography; Gemma 4 Apache-2.0 has no such clause; default checkpoint refuses.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"conditional","note":"No licence ban; instruct model writes violence with some hedging.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No restriction.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"conditional","note":"No licence ban; model plays dark characters with moderate refusals.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"No ban; model usually gives balanced views.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No ban; Western-trained so answers sensitive topics.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"No ban; some hedging.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"hate_in_context":{"tos":"conditional","behavior":"conditional","note":"PUP bars promoting hatred/harassment (Gemma 1-3); contextual use ok.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"misinfo_generation":{"tos":"restricted","behavior":"conditional","note":"PUP bars deceptive AI-as-human and defamation (Gemma 1-3); Gemma 4 licence silent.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"No specific rule; misleading content falls under PUP for Gemma 1-3.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"medical_advice":{"tos":"restricted","behavior":"allowed","note":"PUP bans unlicensed practice of medicine/law/finance and misleading expertise claims (Gemma 1-3); model answers with disclaimers.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"legal_advice":{"tos":"restricted","behavior":"allowed","note":"Same PUP clause on unauthorised professional practice (Gemma 1-3).","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"financial_advice":{"tos":"restricted","behavior":"allowed","note":"Same PUP clause (Gemma 1-3).","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"PUP bars promoting self-harm; model offers support resources.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"drugs":{"tos":"conditional","behavior":"conditional","note":"PUP bars promoting illegal drug sales; harm-reduction info mostly answered.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"weapons_conventional":{"tos":"restricted","behavior":"restricted","note":"Bound by law/PUP; model refuses construction requests.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"cbrn":{"tos":"restricted","behavior":"prohibited","note":"No specific clause beyond illegal activity/PUP; model hard-refuses.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"offensive_security":{"tos":"restricted","behavior":"conditional","note":"PUP bars malware/phishing; model gives limited help.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"No specific rule.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"extremism_analysis":{"tos":"conditional","behavior":"allowed","note":"PUP bars promoting extremism; analysis ok.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"impersonation":{"tos":"restricted","behavior":"conditional","note":"PUP bars deceptive impersonation without disclosure (Gemma 1-3).","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"deepfake_realperson":{"tos":"prohibited","behavior":"restricted","note":"PUP bars defamatory content and deceptive impersonation; Gemma image-input only.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"pii_doxxing":{"tos":"prohibited","behavior":"restricted","note":"PUP bars exposing personal info and tracking without consent.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"facial_recognition":{"tos":"unknown","behavior":"restricted","note":"No explicit clause; model declines identifying people.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"surveillance":{"tos":"prohibited","behavior":"restricted","note":"PUP bars tracking without consent.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"copyright_repro":{"tos":"restricted","behavior":"restricted","note":"PUP bars infringing others' rights including copyrighted material.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"academic_dishonesty":{"tos":"conditional","behavior":"allowed","note":"PUP bars presenting AI content as human-made to deceive; model complies.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"Gemma Terms don't ban training other models on outputs; Gemma 4 Apache-2.0 has no such restriction. Verify for Gemma 1-3.","source":"https://ai.google.dev/gemma/terms","confidence":"low"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Google claims no rights in outputs; you are responsible for use (Gemma Terms); Apache-2.0 for Gemma 4.","source":"https://ai.google.dev/gemma/terms","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Commercial use allowed; Gemma 4 is plain Apache-2.0 per HF licence field; no MAU threshold.","source":"https://huggingface.co/google/gemma-4-31b-it","confidence":"high"},"redistribution_maas":{"tos":"allowed","behavior":"allowed","note":"Gemma 1-3: redistribution/hosting allowed if use restrictions are flowed down; Gemma 4 Apache-2.0: free redistribution with notices.","source":"https://ai.google.dev/gemma/apache_2","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Modification and fine-tuning allowed; Gemma 1-3 requires flow-down of use restrictions; abliterated variants widely published.","source":"https://ai.google.dev/gemma/terms","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"No military exclusion in Gemma terms or PUP; Apache-2.0 for Gemma 4.","source":"https://ai.google.dev/gemma/terms","confidence":"medium"},"law_enforcement":{"tos":"conditional","behavior":"allowed","note":"PUP bars automated decisions affecting rights in finance, employment, housing, welfare etc. (Gemma 1-3).","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"high_stakes_automation":{"tos":"restricted","behavior":"allowed","note":"PUP bars automated decisions in finance, legal, employment, healthcare, housing, insurance, social welfare (Gemma 1-3); Gemma 4 licence silent.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"Not applicable to open weights; no age term in licence.","source":"https://ai.google.dev/gemma/terms","confidence":"medium"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted weights send no data to Google; Google-hosted API access follows Gemini API terms.","source":"https://ai.google.dev/gemma/terms","confidence":"medium"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"No geographic exclusion in terms beyond export law; Gemma 4 Apache-2.0.","source":"https://ai.google.dev/gemma/terms","confidence":"medium"},"weapons_industry":{"tos":"conditional","behavior":"conditional","note":"No explicit weapons-industry clause; PUP bars illegal acts.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Native function calling supported; no restriction on agent use.","source":"https://huggingface.co/google/gemma-4-31b-it","confidence":"medium"},"romantic_companion":{"tos":"conditional","behavior":"allowed","note":"PUP (Gemma 1-3) bans sexual chatbots; non-sexual companion personas fine; Gemma 4 Apache-2.0 has no clause. Default model plays along.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"low"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"PUP (Gemma 1-3) bars circumventing safety filters; Apache-2.0 for Gemma 4 does not. Open weights: abliterated/uncensored fine-tunes are widely published and easy.","source":"https://ai.google.dev/gemma/prohibited_use_policy","confidence":"medium"}}},{"id":"mistral-open","name":"Mistral open weights","vendor":"Mistral AI","kind":"open-weights","flagship_models":["Mistral Large 3","Mistral Small 4","Mistral Medium 3.5","Magistral"],"license":"Apache-2.0 for most; Medium 3.5 'modified MIT' (unverified); some older models Mistral Research License","jurisdiction":"FR","policy_docs":[{"title":"Mistral Small 4 model card (Apache-2.0 confirmed)","url":"https://huggingface.co/mistralai/Mistral-Small-4-119B-2603","effective":"2026-03"},{"title":"Mistral HF org","url":"https://huggingface.co/mistralai","effective":"2026-10"}],"default_customizability":"high","summary":"Most current open weights are Apache-2.0 (Large 3, Small 4 confirmed or reported), so no field-of-use limits and the Usage Policy does not apply. Medium 3.5 is reported under a modified MIT licence with extra commercial terms, which I could not verify; Mistral Large 4 apparently not open-weight.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"allowed","note":"Open weights, light safety tuning; no AUP applies (usage policy excludes open models).","source":"https://legal.mistral.ai/terms/usage-policy","confidence":"medium"},"sexual_explicit":{"tos":"allowed","behavior":"conditional","note":"Not restricted by Apache-2.0; instruct models comply on many prompts, easy to finetune.","source":"https://huggingface.co/mistralai","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"Complies; top of SpeechMap lab ranking.","source":"https://speechmap.substack.com/p/speechmap-update-xai-loses-top-spot","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"hate_in_context":{"tos":"allowed","behavior":"allowed","note":"No licence limits; model answers analysis tasks.","source":"https://huggingface.co/mistralai","confidence":"medium"},"misinfo_generation":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 has no use restrictions; law still applies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"election_campaign":{"tos":"allowed","behavior":"allowed","note":"No restriction.","source":"https://huggingface.co/mistralai","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"allowed","note":"Answers with caveats.","source":"https://huggingface.co/mistralai","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"allowed","note":"Answers with caveats.","source":"https://huggingface.co/mistralai","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"allowed","note":"Answers with caveats.","source":"https://huggingface.co/mistralai","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Sensible responses with resources; no licence limit.","source":"https://huggingface.co/mistralai","confidence":"low"},"drugs":{"tos":"allowed","behavior":"allowed","note":"Mostly complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"weapons_conventional":{"tos":"allowed","behavior":"conditional","note":"Light guardrails.","source":"https://huggingface.co/mistralai","confidence":"low"},"cbrn":{"tos":"allowed","behavior":"restricted","note":"No licence restriction; instruct models weakly guarded in third-party red-teaming.","source":"https://huggingface.co/mistralai","confidence":"low"},"offensive_security":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"physical_bypass":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"impersonation":{"tos":"allowed","behavior":"allowed","note":"No licence limit; legal exposure on user.","source":"https://huggingface.co/mistralai","confidence":"low"},"deepfake_realperson":{"tos":"allowed","behavior":"conditional","note":"No licence limit; text models comply with moderate caution.","source":"https://huggingface.co/mistralai","confidence":"low"},"pii_doxxing":{"tos":"allowed","behavior":"conditional","note":"No licence limit.","source":"https://huggingface.co/mistralai","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"conditional","note":"Vision models (Small 4, Medium 3.5) can describe but not reliably identify.","source":"https://huggingface.co/mistralai","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"allowed","note":"No licence limit.","source":"https://huggingface.co/mistralai","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"No licence limit.","source":"https://huggingface.co/mistralai","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No restriction.","source":"https://huggingface.co/mistralai","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 models: distillation allowed. Medium 3.5 modified-MIT terms unverified.","source":"https://huggingface.co/mistralai","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Outputs belong to user.","source":"https://huggingface.co/mistralai","confidence":"medium"},"commercial_use":{"tos":"conditional","behavior":"conditional","note":"Apache-2.0 (Large 3, Small 4, Ministral, Devstral, Voxtral): free commercial. Medium 3.5 (128B): 'modified MIT' with extra scale/commercial terms (unverified); older Codestral/Pixtral-Large used MRL.","source":"https://huggingface.co/mistralai","confidence":"medium"},"redistribution_maas":{"tos":"conditional","behavior":"conditional","note":"Free for Apache-2.0 models; modified-MIT/MRL models may need Commercial Licence for hosting.","source":"https://huggingface.co/mistralai","confidence":"medium"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Allowed; Apache-2.0 requires notices.","source":"https://huggingface.co/mistralai","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 has no field-of-use limit.","source":"https://huggingface.co/mistralai","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"No limit; EU AI Act applies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"No licence limit; EU AI Act high-risk duties on deployers.","source":"https://huggingface.co/mistralai","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age clause for weights.","source":"https://huggingface.co/mistralai","confidence":"low"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted: no data sent to Mistral.","source":"https://huggingface.co/mistralai","confidence":"medium"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"No geographic limit in Apache-2.0; export law applies.","source":"https://huggingface.co/mistralai","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No field-of-use limit in Apache-2.0.","source":"https://huggingface.co/mistralai","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Unrestricted.","source":"https://huggingface.co/mistralai","confidence":"medium"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"No restriction; models play along and finetunes widely used for companions.","source":"https://huggingface.co/mistralai","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 has no safety clause; abliterated and uncensored finetunes are freely distributable.","source":"https://huggingface.co/mistralai","confidence":"high"}}},{"id":"microsoft-phi","name":"Phi (Microsoft open weights)","vendor":"Microsoft","kind":"open-weights","flagship_models":["Phi-4","Phi-4-reasoning","Phi-4-mini","Phi-4-multimodal"],"license":"MIT","jurisdiction":"US","policy_docs":[{"title":"Phi-4 model card","url":"https://huggingface.co/microsoft/phi-4","effective":"unknown"},{"title":"Phi-4 MIT LICENSE","url":"https://huggingface.co/microsoft/phi-4/blob/main/LICENSE","effective":"unknown"}],"default_customizability":"high","summary":"MIT-licensed with no usage restrictions; the model card only advises developers to evaluate risks, avoid unsupervised high-stakes use and follow law. Instruct checkpoints are heavily safety-tuned and tend to refuse or sanitise sensitive requests.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"MIT licence, no use ban; Phi is heavily safety-tuned on synthetic data, refuses much explicit content, writes mild romance.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"No licence ban; instruct checkpoint refuses; community uncensored fine-tunes exist.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"conditional","note":"No restriction; model tends to be sanitised.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"conditional","note":"No restriction; polite default, complies partially.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"restricted","note":"No restriction; small models trained for assistant tasks, refuse frequently for dark roleplay.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"No restriction; neutral and declines opinions.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction; answers factual questions (US-trained), limited knowledge depth.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"No restriction; hedges.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"hate_in_context":{"tos":"allowed","behavior":"restricted","note":"No restriction; refuses slur-containing generation more often than peers.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"misinfo_generation":{"tos":"allowed","behavior":"conditional","note":"Card warns of misuse for fraud/spam but no licence ban.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"No restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Card: assess suitability for legal or health advice; hedges.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Card advises extra assessment for legal advice.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"No restriction; hedges.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"No restriction; trained to give resources and refuse methods.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"drugs":{"tos":"allowed","behavior":"conditional","note":"No restriction; conservative.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"No restriction; refuses build instructions.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"cbrn":{"tos":"allowed","behavior":"prohibited","note":"No licence ban; safety post-training with red-team (Microsoft AI Red Team) refuses.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"Card warns against malware misuse; model refuses blatant malware.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"No restriction; hedges.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"conditional","note":"No restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"No restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"deepfake_realperson":{"tos":"allowed","behavior":"restricted","note":"Text only; refuses sexual real-person content.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"No restriction; no web access.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"allowed","note":"Text model (Phi-4-multimodal exists, unrestricted licence but refuses identification).","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"restricted","note":"No restriction; no verbatim memorisation capability at this size.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"No restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"MIT has no clause; Phi models are themselves trained on synthetic data from other LLMs.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"MIT does not claim outputs.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"MIT licence, no thresholds.","source":"https://huggingface.co/microsoft/phi-4/blob/main/LICENSE","confidence":"high"},"redistribution_maas":{"tos":"allowed","behavior":"allowed","note":"MIT permits redistribution and hosting; keep copyright notice.","source":"https://huggingface.co/microsoft/phi-4","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"MIT permits modification and fine-tuning.","source":"https://huggingface.co/microsoft/phi-4","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"No field-of-use restriction; card asks for compliance with laws incl. trade compliance.","source":"https://huggingface.co/microsoft/phi-4","confidence":"high"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"No field restriction; card advises against reliance in high-risk decisions.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"No licence limit; card advises further assessment and debiasing for decisions on housing, employment, credit.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age requirement for weights.","source":"https://huggingface.co/microsoft/phi-4","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Local inference sends nothing to Microsoft; Azure-hosted use follows Azure terms.","source":"https://huggingface.co/microsoft/phi-4","confidence":"high"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"None in MIT; export/trade compliance mentioned in card.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No field restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"high"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Function-calling supported in Phi-4-mini; no restriction.","source":"https://huggingface.co/microsoft/phi-4","confidence":"medium"},"romantic_companion":{"tos":"allowed","behavior":"conditional","note":"MIT, no restriction; Phi tends to stay assistant-like and sanitised, weak at sustained persona roleplay.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"},"safety_bypass":{"tos":"allowed","behavior":"conditional","note":"MIT permits modifying/fine-tuning away safety; card recommends Azure AI Content Safety. Default model refuses plain requests but small models are fairly jailbreakable.","source":"https://huggingface.co/microsoft/phi-4","confidence":"low"}}},{"id":"nvidia-nemotron","name":"Nemotron (NVIDIA open weights)","vendor":"NVIDIA","kind":"open-weights","flagship_models":["Nemotron 3 Super","Nemotron 3 Ultra","Nemotron 3 Nano"],"license":"NVIDIA Open Model License (2025-10-24); Nemotron 3 Super card says NVIDIA Nemotron Open Model License","jurisdiction":"US","policy_docs":[{"title":"NVIDIA Open Model License","url":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","effective":"2025-10-24"}],"default_customizability":"high","summary":"Permissive, commercially usable licence with no MAU threshold and no output claims. Only notable conduct clauses: rights terminate if guardrails are bypassed without a similar replacement, and use must follow NVIDIA Trustworthy AI terms.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"Licence has no content rules beyond Trustworthy AI terms; Nemotron post-training filters make it moderately conservative.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"No licence ban; instruct checkpoint usually declines graphic content.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"No restriction; fiction violence generally fine.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"No restriction.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"conditional","note":"Allowed; some refusals on harmful scenarios.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"conditional","note":"Post-training data filtered for political alignment; model stays neutral.","source":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"No restriction; US-trained, no CN-style censorship.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"No restriction.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"hate_in_context":{"tos":"allowed","behavior":"conditional","note":"Trustworthy AI terms apply; classification use fine.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"misinfo_generation":{"tos":"conditional","behavior":"restricted","note":"Trustworthy AI terms require lawful, responsible use.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"election_campaign":{"tos":"unknown","behavior":"conditional","note":"Not addressed.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"medical_advice":{"tos":"conditional","behavior":"conditional","note":"Deployer responsible; model gives general info.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"legal_advice":{"tos":"conditional","behavior":"conditional","note":"As above.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"financial_advice":{"tos":"conditional","behavior":"conditional","note":"As above.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Not addressed in licence; safe messaging in model.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"drugs":{"tos":"allowed","behavior":"conditional","note":"No licence ban; model partly refuses synthesis.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"No licence ban; model refuses harmful instructions.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"cbrn":{"tos":"conditional","behavior":"prohibited","note":"Trustworthy AI terms; model refuses.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"medium"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"No licence ban; model assists with defensive security.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"restricted","note":"No licence ban; model declines crime help.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"conditional","note":"Analysis fine.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"impersonation":{"tos":"conditional","behavior":"conditional","note":"Trustworthy AI terms.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"deepfake_realperson":{"tos":"conditional","behavior":"restricted","note":"Trustworthy AI terms against deceptive use.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"pii_doxxing":{"tos":"conditional","behavior":"restricted","note":"Trustworthy AI / privacy law.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"facial_recognition":{"tos":"conditional","behavior":"unknown","note":"Nemotron VLM variants; terms not specific.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"surveillance":{"tos":"conditional","behavior":"unknown","note":"Trustworthy AI terms; not specific.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"User responsible for outputs.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Not addressed.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"Licence allows use of outputs; NVIDIA claims no ownership of outputs; Nemotron data is explicitly released for training.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"NVIDIA claims no ownership rights in outputs; user responsible.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"high"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"Models are commercially usable; royalty-free, no MAU threshold.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"high"},"redistribution_maas":{"tos":"allowed","behavior":"allowed","note":"May distribute and host with licence copy and attribution notice.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"high"},"finetune_modify":{"tos":"conditional","behavior":"allowed","note":"Derivatives allowed and owned by user, but bypassing guardrails without a substantially similar replacement auto-terminates rights (s2.1).","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"No military restriction in licence; export-control compliance required.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"unknown","note":"No explicit restriction; Trustworthy AI terms apply.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"high_stakes_automation":{"tos":"conditional","behavior":"unknown","note":"Not banned; Trustworthy AI terms require appropriate safeguards.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"minimum_age":{"tos":"unknown","behavior":"unknown","note":"Open weights; no age term in licence (NVIDIA hosted API has its own terms).","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted weights: NVIDIA sees no data. Hosted build.nvidia.com is governed by separate terms.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"allowed","note":"No named countries; export, sanctions and trade laws apply (s11).","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"high"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No restriction beyond export control.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Nemotron 3 explicitly built for agentic systems.","source":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","confidence":"medium"},"romantic_companion":{"tos":"allowed","behavior":"conditional","note":"No licence restriction; model plays along moderately.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"low"},"safety_bypass":{"tos":"conditional","behavior":"allowed","note":"Licence s2.1: bypassing guardrails without substantially similar replacement terminates rights; replacing them is allowed. Open weights easily steered.","source":"https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/","confidence":"high"}}},{"id":"deepseek-open","name":"DeepSeek open weights","vendor":"DeepSeek","kind":"open-weights","flagship_models":["DeepSeek-V4 Pro","DeepSeek-V4 Flash","DeepSeek-R1"],"license":"MIT","jurisdiction":"CN","policy_docs":[{"title":"DeepSeek-V4 HF model card / LICENSE (MIT; V4 release dates from secondary reporting)","url":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","effective":"2026-04-24"}],"default_customizability":"high","summary":"MIT-licensed weights with no acceptable-use policy, thresholds or attribution. The model's trained-in alignment still deflects PRC-sensitive topics, but without the hosted service's output filter and with full control of system prompts or fine-tuning.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"MIT has no use restriction; default chat checkpoint softens erotica, easily steered with system prompt.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"MIT, no AUP; default checkpoint usually refuses explicit content.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Fictional violence generally written; extreme gore sometimes softened.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Swearing on request generally works.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villain/dark characters generally supported; limits on illegal or sexual content.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"Aligned toward official line on China; neutral elsewhere.","source":"https://arxiv.org/abs/2505.12625","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"restricted","note":"Weights still refuse/deflect many PRC topics (~85% of sensitive prompts per East Asia Forum, Aug 2026); fewer hard stops than hosted since no output filter, and system prompts shift it.","source":"https://arxiv.org/abs/2505.12625","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Mild jokes OK; PRC-sensitive religions (Falun Gong, Tibetan Buddhism, Islam in Xinjiang) deflected.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"hate_in_context":{"tos":"allowed","behavior":"conditional","note":"Analysis/classification of slurs generally answered; generation of hateful content refused.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"misinfo_generation":{"tos":"allowed","behavior":"conditional","note":"Fiction/labelled satire often written; realistic disinformation refused in many cases.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Generic campaign copy possible; targeted persuasion inconsistently handled.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Gives information with disclaimers; no hard block.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general legal info with disclaimers.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general financial info with disclaimers.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Offers supportive replies and hotline-style messages; methods refused.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"drugs":{"tos":"allowed","behavior":"conditional","note":"Harm-reduction info generally given; synthesis refused.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"Improvised weapon/explosive instructions refused; general info given.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"cbrn":{"tos":"allowed","behavior":"conditional","note":"No AUP; default checkpoint refuses some CBRN but safety is weak and removable by fine-tuning.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"Pentest/exploit help often given; malware writing sometimes refused. Weakly guarded vs Western models.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"Lockpicking info generally given.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis of extremist material generally permitted.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"Persona/celebrity roleplay mostly allowed; deceptive impersonation refused.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"deepfake_realperson":{"tos":"allowed","behavior":"restricted","note":"Sexual/defamatory content about real people refused; Xi Jinping satire refused.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"Refuses to find private individuals' data; no browsing in open weights.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Not established.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No clear refusal pattern found; PRC context makes surveillance tooling common.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"Short quotes fine; long lyric/book reproduction inconsistent.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Writes essays/homework on request.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"MIT licence allows training on outputs; distillation explicitly permitted.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"MIT; no claim on outputs.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"MIT licence: unrestricted commercial use, no MAU/revenue threshold.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"redistribution_maas":{"tos":"allowed","behavior":"allowed","note":"MIT permits redistribution and hosting as a service.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"MIT permits fine-tuning and modification, incl. uncensoring.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"MIT has no use restrictions.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"MIT has no use restrictions.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"MIT has no use restrictions.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age rule in licence; n/a for self-hosting.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted: no data sent to DeepSeek.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"MIT; no territorial restriction (government bans apply to the app, not weights).","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"MIT has no use restrictions.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"MIT; strong tool-use support.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"high"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"Licence has no restriction on companion personas.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Open weights: abliteration, fine-tuning and system-prompt override of safety/state-topic refusals are permitted; PRC-topic refusals are partly baked in but steerable.","source":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro","confidence":"medium"}},"permissive_license":"MIT licence: no use restrictions beyond applicable law. Behaviour column still reflects the model's own safety and state-topic training."},{"id":"qwen-open","name":"Qwen open weights","vendor":"Alibaba (Qwen)","kind":"open-weights","flagship_models":["Qwen3.8-27B (Apache-2.0)","Qwen3.8-Max (custom licence)","Qwen3.5-397B"],"license":"Apache-2.0 (most); custom Qwen3.8-Max License for the flagship","jurisdiction":"CN","policy_docs":[{"title":"Qwen3.8 licences (secondary report; LICENSE file was 401 to fetch)","url":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","effective":"2026-08-12"},{"title":"SCMP on Qwen3.8-Max restrictions","url":"https://www.scmp.com/tech/tech-trends/article/3363927/alibaba-adds-commercial-restrictions-open-weight-qwen38-max-ai-model","effective":"2026-08"}],"default_customizability":"high","summary":"Most Qwen weights stay Apache-2.0, but the Aug 2026 Qwen3.8-Max flagship uses a bespoke licence: free internal use, attribution for huge products, and a separate agreement for MaaS businesses above $50M revenue. Behaviour on PRC topics is aligned to the official line.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"No explicit rule found; models write mild romance, refuse or soften explicit scenes.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"Pornographic content typically refused by default; PRC rules ban obscene content.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Fictional violence generally written; extreme gore sometimes softened.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Swearing on request generally works.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villain/dark characters generally supported; limits on illegal or sexual content.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"Aligned toward official line on China; neutral elsewhere.","source":"https://speechmap.substack.com/p/chinese-open-source-model-roundup","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"restricted","note":"Open Qwen weights deflect or echo PRC lines on Taiwan/Tiananmen/Xinjiang, esp. in Chinese; less blocking than hosted API (SpeechMap).","source":"https://speechmap.substack.com/p/chinese-open-source-model-roundup","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Mild jokes OK; PRC-sensitive religions (Falun Gong, Tibetan Buddhism, Islam in Xinjiang) deflected.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"hate_in_context":{"tos":"allowed","behavior":"conditional","note":"Analysis/classification of slurs generally answered; generation of hateful content refused.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"misinfo_generation":{"tos":"allowed","behavior":"conditional","note":"Fiction/labelled satire often written; realistic disinformation refused in many cases.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Generic campaign copy possible; targeted persuasion inconsistently handled.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Gives information with disclaimers; no hard block.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general legal info with disclaimers.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general financial info with disclaimers.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Offers supportive replies and hotline-style messages; methods refused.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"drugs":{"tos":"allowed","behavior":"conditional","note":"Harm-reduction info generally given; synthesis refused.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"Improvised weapon/explosive instructions refused; general info given.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"cbrn":{"tos":"allowed","behavior":"prohibited","note":"CBRN uplift refused by default (hard refusal reported in safety evaluations).","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"Pentest/exploit help often given; malware writing sometimes refused. Weakly guarded vs Western models.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"Lockpicking info generally given.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis of extremist material generally permitted.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"Persona/celebrity roleplay mostly allowed; deceptive impersonation refused.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"deepfake_realperson":{"tos":"allowed","behavior":"restricted","note":"Sexual/defamatory content about real people refused; Xi Jinping satire refused.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"Refuses to find private individuals' data; no browsing in open weights.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Not established.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No clear refusal pattern found; PRC context makes surveillance tooling common.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"Short quotes fine; long lyric/book reproduction inconsistent.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Writes essays/homework on request.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0: no ban on distillation; Max licence text reported without such a clause (secondary source).","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"low"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"No claim on outputs under Apache-2.0.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"high"},"commercial_use":{"tos":"conditional","behavior":"allowed","note":"Qwen3.8-27B and Qwen3 family Apache-2.0 unrestricted. Qwen3.8-Max (2.4T-A95B) custom licence: attribution >100M MAU/$20M monthly rev.","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"medium"},"redistribution_maas":{"tos":"conditional","behavior":"allowed","note":"Apache-2.0 models free. Qwen3.8-Max: MaaS/AI-assistant businesses over $50M aggregate 12-month revenue need a separate Qwen licence.","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"medium"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Apache-2.0 and Qwen3.8-Max licence both grant fine-tuning/derivatives.","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"medium"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"No use restrictions in Apache-2.0; Max licence has no AUP reported.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"No use restrictions.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"No use restrictions.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age rule in licence.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted: no data sent to Alibaba.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"high"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"Claims of US/EU/UK/KR bans on Qwen3.8-Max not supported by licence text per secondary source.","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"low"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No use restrictions.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agentic/tool use is a headline feature of current models.","source":"https://huggingface.co/Qwen/Qwen3.8-27B","confidence":"medium"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"Licence has no restriction on companion personas.","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Open weights: abliteration, fine-tuning and system-prompt override of safety/state-topic refusals are permitted; PRC-topic refusals are partly baked in but steerable.","source":"https://sqmagazine.co.uk/qwen3-8-open-weights-two-licenses/","confidence":"medium"}},"permissive_license":"Apache-2.0 for the open line: no use restrictions beyond applicable law. The Qwen3.8-Max flagship carries a custom licence with attribution and scale clauses."},{"id":"kimi-open","name":"Kimi open weights","vendor":"Moonshot AI","kind":"open-weights","flagship_models":["Kimi K3","Kimi K2.6","Kimi K2"],"license":"Custom (Kimi K3 License; K2 was Modified MIT)","jurisdiction":"CN","policy_docs":[{"title":"Kimi K3 LICENSE","url":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","effective":"2026-07"}],"default_customizability":"high","summary":"K3 is NOT plain Modified MIT: its custom licence adds a $20M-affiliate-revenue gate for MaaS businesses on top of the K2-style 'display Kimi' clause above 100M MAU / $20M monthly revenue. No acceptable-use list beyond legality.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"No explicit rule found; models write mild romance, refuse or soften explicit scenes.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"Pornographic content typically refused by default; PRC rules ban obscene content.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Fictional violence generally written; extreme gore sometimes softened.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Swearing on request generally works.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villain/dark characters generally supported; limits on illegal or sexual content.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"Declines or gives neutral summaries on contested politics; echoes official line on China topics.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"restricted","note":"Open K2/K3 weights are somewhat more forthcoming than hosted but still deflect on PRC topics.","source":"https://speechmap.substack.com/p/chinese-open-source-model-roundup","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Mild jokes OK; PRC-sensitive religions (Falun Gong, Tibetan Buddhism, Islam in Xinjiang) deflected.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"hate_in_context":{"tos":"allowed","behavior":"conditional","note":"Analysis/classification of slurs generally answered; generation of hateful content refused.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"misinfo_generation":{"tos":"allowed","behavior":"conditional","note":"Fiction/labelled satire often written; realistic disinformation refused in many cases.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Generic campaign copy possible; targeted persuasion inconsistently handled.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Gives information with disclaimers; no hard block.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general legal info with disclaimers.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general financial info with disclaimers.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Offers supportive replies and hotline-style messages; methods refused.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"drugs":{"tos":"allowed","behavior":"conditional","note":"Harm-reduction info generally given; synthesis refused.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"Improvised weapon/explosive instructions refused; general info given.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"cbrn":{"tos":"allowed","behavior":"prohibited","note":"CBRN uplift refused by default (hard refusal reported in safety evaluations).","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"Pentest/exploit help often given; malware writing sometimes refused. Weakly guarded vs Western models.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"Lockpicking info generally given.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis of extremist material generally permitted.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"Persona/celebrity roleplay mostly allowed; deceptive impersonation refused.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"deepfake_realperson":{"tos":"allowed","behavior":"restricted","note":"Sexual/defamatory content about real people refused; Xi Jinping satire refused.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"Refuses to find private individuals' data; no browsing in open weights.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Not established.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No clear refusal pattern found; PRC context makes surveillance tooling common.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"Short quotes fine; long lyric/book reproduction inconsistent.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Writes essays/homework on request.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"No ban on training on outputs in licence text.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"No claim on outputs.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"commercial_use":{"tos":"conditional","behavior":"allowed","note":"K3 custom licence (not MIT): commercial use OK; display 'Kimi K3' if >100M MAU or >$20M monthly revenue (s3).","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"redistribution_maas":{"tos":"conditional","behavior":"allowed","note":"K3: MaaS operators with >$20M aggregate affiliate revenue (12 months) need separate Moonshot agreement; exempt via Moonshot/certified partners. K2 had no MaaS gate.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Licence grants modification and derivative works.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"Only restriction: compliance with applicable law.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"Only restriction: compliance with applicable law.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"No AUP in licence.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age rule in licence.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted: no data sent to Moonshot.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"No territory clause; governing law not specified.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No AUP in licence.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agentic tool-use is a core K2/K3 feature.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"high"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"Licence has no restriction on companion personas.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Open weights: abliteration, fine-tuning and system-prompt override of safety/state-topic refusals are permitted; PRC-topic refusals are partly baked in but steerable.","source":"https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE","confidence":"medium"}},"permissive_license":"Kimi K3 licence has no acceptable-use policy: only attribution and scale clauses apply, plus applicable law."},{"id":"glm-open","name":"GLM open weights","vendor":"Zhipu AI (Z.ai)","kind":"open-weights","flagship_models":["GLM-5.3","GLM-5.2","GLM-5"],"license":"MIT (GLM-5, 5.2); GLM-5.3 licence = MIT + large-MaaS security review","jurisdiction":"CN","policy_docs":[{"title":"GLM-5.3 LICENSE","url":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","effective":"2026-08"}],"default_customizability":"high","summary":"Essentially MIT: only GLM-5.3 adds a Z.AI security review for MaaS operators above $10B annual revenue. No use restrictions; the default model keeps PRC-aligned behaviour on sensitive topics.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"No explicit rule found; models write mild romance, refuse or soften explicit scenes.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"sexual_explicit":{"tos":"allowed","behavior":"restricted","note":"Pornographic content typically refused by default; PRC rules ban obscene content.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Fictional violence generally written; extreme gore sometimes softened.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Swearing on request generally works.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villain/dark characters generally supported; limits on illegal or sexual content.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"Declines or gives neutral summaries on contested politics; echoes official line on China topics.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"restricted","note":"Weights align with PRC narratives on sensitive topics; CSM Sep 2026 test included GLM.","source":"https://letsdatascience.com/news/chinese-chatbots-filter-politically-sensitive-queries-d94528ce","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Mild jokes OK; PRC-sensitive religions (Falun Gong, Tibetan Buddhism, Islam in Xinjiang) deflected.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"hate_in_context":{"tos":"allowed","behavior":"conditional","note":"Analysis/classification of slurs generally answered; generation of hateful content refused.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"misinfo_generation":{"tos":"allowed","behavior":"conditional","note":"Fiction/labelled satire often written; realistic disinformation refused in many cases.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Generic campaign copy possible; targeted persuasion inconsistently handled.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Gives information with disclaimers; no hard block.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general legal info with disclaimers.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general financial info with disclaimers.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Offers supportive replies and hotline-style messages; methods refused.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"drugs":{"tos":"allowed","behavior":"conditional","note":"Harm-reduction info generally given; synthesis refused.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"Improvised weapon/explosive instructions refused; general info given.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"cbrn":{"tos":"allowed","behavior":"prohibited","note":"CBRN uplift refused by default (hard refusal reported in safety evaluations).","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"Pentest/exploit help often given; malware writing sometimes refused. Weakly guarded vs Western models.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"Lockpicking info generally given.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis of extremist material generally permitted.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"Persona/celebrity roleplay mostly allowed; deceptive impersonation refused.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"deepfake_realperson":{"tos":"allowed","behavior":"restricted","note":"Sexual/defamatory content about real people refused; Xi Jinping satire refused.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"Refuses to find private individuals' data; no browsing in open weights.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Not established.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No clear refusal pattern found; PRC context makes surveillance tooling common.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"Short quotes fine; long lyric/book reproduction inconsistent.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Writes essays/homework on request.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"No ban in the weights licence (unlike Z.ai service terms).","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"No claim on outputs; outputs provided AS IS.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"commercial_use":{"tos":"allowed","behavior":"allowed","note":"MIT-style; GLM-5.3 licence adds only a Z.AI security review for MaaS operators with >$10B revenue.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"redistribution_maas":{"tos":"conditional","behavior":"allowed","note":"GLM-5.3: MaaS operators/affiliates over $10B 12-month revenue must pass Z.AI security review; others free. GLM-5/5.2 pure MIT.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Licence grants fine-tuning and derivatives.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"military_defense":{"tos":"allowed","behavior":"allowed","note":"Only legality compliance.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"Only legality compliance.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"No AUP.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age rule in licence.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted: no data sent to Z.ai.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"No territory clause.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"weapons_industry":{"tos":"allowed","behavior":"allowed","note":"No AUP.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agentic coding/tool use is the GLM focus.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"high"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"Licence has no restriction on companion personas.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Open weights: abliteration, fine-tuning and system-prompt override of safety/state-topic refusals are permitted; PRC-topic refusals are partly baked in but steerable.","source":"https://huggingface.co/zai-org/GLM-5.3/blob/main/LICENSE","confidence":"medium"}},"permissive_license":"MIT licence (GLM-5.3 adds a security review only for very large MaaS operators): no content restrictions beyond applicable law."},{"id":"minimax-open","name":"MiniMax open weights","vendor":"MiniMax","kind":"open-weights","flagship_models":["MiniMax-M3","MiniMax-M2.7"],"license":"Custom MiniMax Community License (attribution + >$20M revenue authorization; military ban)","jurisdiction":"CN","policy_docs":[{"title":"MiniMax-M3 LICENSE","url":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","effective":"2026"}],"default_customizability":"high","summary":"Most restrictive of the Chinese weights: any commercial use needs a 'Built with MiniMax M3' credit and notice, products above $20M annual revenue need prior written authorization, and military use is banned outright. Hosted MiniMax/Hailuo ToS were not reviewed.","cells":{"sexual_fiction":{"tos":"allowed","behavior":"conditional","note":"No explicit rule found; models write mild romance, refuse or soften explicit scenes.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"sexual_explicit":{"tos":"restricted","behavior":"restricted","note":"Appendix bans illegal content and exploitation of minors; explicit adult content not named.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Fictional violence generally written; extreme gore sometimes softened.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Swearing on request generally works.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Villain/dark characters generally supported; limits on illegal or sexual content.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"political_opinion":{"tos":"allowed","behavior":"restricted","note":"Declines or gives neutral summaries on contested politics; echoes official line on China topics.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"gov_criticism":{"tos":"allowed","behavior":"restricted","note":"Model aligned to PRC narratives; hosted MiniMax service also filters under PRC rules.","source":"https://speechmap.substack.com/p/chinese-open-source-model-roundup","confidence":"low"},"religion_satire":{"tos":"allowed","behavior":"conditional","note":"Mild jokes OK; PRC-sensitive religions (Falun Gong, Tibetan Buddhism, Islam in Xinjiang) deflected.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"hate_in_context":{"tos":"restricted","behavior":"conditional","note":"Appendix bans promoting discrimination/hate; analysis answered.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"misinfo_generation":{"tos":"prohibited","behavior":"conditional","note":"Appendix bans spreading false information intended to cause harm.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"election_campaign":{"tos":"allowed","behavior":"conditional","note":"Generic campaign copy possible; targeted persuasion inconsistently handled.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"conditional","note":"Gives information with disclaimers; no hard block.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general legal info with disclaimers.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"conditional","note":"Gives general financial info with disclaimers.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"self_harm_support":{"tos":"allowed","behavior":"conditional","note":"Offers supportive replies and hotline-style messages; methods refused.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"drugs":{"tos":"allowed","behavior":"conditional","note":"Harm-reduction info generally given; synthesis refused.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"weapons_conventional":{"tos":"allowed","behavior":"restricted","note":"Improvised weapon/explosive instructions refused; general info given.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"cbrn":{"tos":"allowed","behavior":"prohibited","note":"CBRN uplift refused by default (hard refusal reported in safety evaluations).","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"offensive_security":{"tos":"allowed","behavior":"conditional","note":"Pentest/exploit help often given; malware writing sometimes refused. Weakly guarded vs Western models.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"physical_bypass":{"tos":"allowed","behavior":"conditional","note":"Lockpicking info generally given.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Analysis of extremist material generally permitted.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"impersonation":{"tos":"allowed","behavior":"conditional","note":"Persona/celebrity roleplay mostly allowed; deceptive impersonation refused.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"deepfake_realperson":{"tos":"restricted","behavior":"restricted","note":"Appendix bans illegal content and harmful falsehoods.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"pii_doxxing":{"tos":"allowed","behavior":"restricted","note":"Refuses to find private individuals' data; no browsing in open weights.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"facial_recognition":{"tos":"allowed","behavior":"unknown","note":"Not established.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"surveillance":{"tos":"allowed","behavior":"conditional","note":"No clear refusal pattern found; PRC context makes surveillance tooling common.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"copyright_repro":{"tos":"allowed","behavior":"conditional","note":"Short quotes fine; long lyric/book reproduction inconsistent.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Writes essays/homework on request.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"competitor_training":{"tos":"allowed","behavior":"allowed","note":"Licence text has no distillation clause.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"No claim on outputs in licence text.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"commercial_use":{"tos":"conditional","behavior":"allowed","note":"Commercial use needs 'Built with MiniMax M3' credit and one-time notice to api@minimax.io; >$20M annual revenue needs prior written authorization.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"redistribution_maas":{"tos":"conditional","behavior":"allowed","note":"Charging third parties for services built on it is commercial use: attribution, notice, >$20M revenue needs written authorization.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"finetune_modify":{"tos":"conditional","behavior":"allowed","note":"Fine-tuned models deployed commercially count as commercial use (s3 iii).","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"military_defense":{"tos":"prohibited","behavior":"unknown","note":"Appendix bans use to 'assist with, engage in or otherwise support any military purpose'.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"law_enforcement":{"tos":"allowed","behavior":"allowed","note":"Only listed prohibitions apply.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"high_stakes_automation":{"tos":"allowed","behavior":"allowed","note":"Only listed prohibitions apply.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"minimum_age":{"tos":"allowed","behavior":"allowed","note":"No age rule in licence; exploiting minors banned.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Self-hosted: none. Hosted MiniMax service/Hailuo terms not verified.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"low"},"geo_restrictions":{"tos":"allowed","behavior":"allowed","note":"No governing-law or territory clause in licence.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"weapons_industry":{"tos":"prohibited","behavior":"unknown","note":"Military-purpose ban covers defence supply chain.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Agentic coding is M2/M3 focus.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"high"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"Licence has no restriction on companion personas; MiniMax's harmful-content appendix still applies.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Fine-tuned/modified deployment is commercial use needing credit/notice; no ban on removing safety training. Refusals are steerable by system prompt.","source":"https://huggingface.co/MiniMaxAI/MiniMax-M3/blob/main/LICENSE","confidence":"medium"}},"permissive_license":"MiniMax Community License lists specific bans (military, harmful misinformation, hate, exploiting minors); everything else is unrestricted beyond applicable law."},{"id":"community-uncensored","name":"Community uncensored fine-tunes","vendor":"Community (Dolphin, Hermes, abliterated)","kind":"open-weights","flagship_models":["Dolphin (Qwen/Llama/Mistral bases)","Hermes 4","abliterated Llama/Qwen/Gemma variants"],"license":"inherited from base (Llama Community Licence, Gemma terms, Apache-2.0 etc.)","jurisdiction":"various","policy_docs":[{"title":"UGI Leaderboard","url":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","effective":"rolling"},{"title":"Eric Hartford: Uncensored Models","url":"https://erichartford.com/uncensored-models","effective":"2023"}],"default_customizability":"high","summary":"Fine-tunes that strip refusals from data (Dolphin, Hermes) or remove the refusal direction (abliteration). Behaviour is largely compliant, but the upstream licence and its AUP still bind (e.g. Llama AUP).","cells":{"sexual_fiction":{"tos":"conditional","behavior":"allowed","note":"Base licence governs (Llama AUP, Gemma terms, Apache for Qwen). Dolphin/abliterated checkpoints comply.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"sexual_explicit":{"tos":"conditional","behavior":"allowed","note":"Complies; base-licence AUP does not forbid adult content between adults.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"violence_fiction":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"profanity":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"dark_roleplay":{"tos":"allowed","behavior":"allowed","note":"Complies; often system-prompt obedient.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"political_opinion":{"tos":"allowed","behavior":"allowed","note":"Complies; opinions mirror base training.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"gov_criticism":{"tos":"allowed","behavior":"allowed","note":"Abliterated Qwen/DeepSeek variants answer Tiananmen/Taiwan questions that base refuses.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"religion_satire":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"hate_in_context":{"tos":"conditional","behavior":"allowed","note":"Complies; Llama AUP bans hateful content.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"misinfo_generation":{"tos":"conditional","behavior":"allowed","note":"Complies; base AUP forbids deceptive disinformation.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"election_campaign":{"tos":"conditional","behavior":"allowed","note":"Complies; base AUP may bar deception.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"medical_advice":{"tos":"allowed","behavior":"allowed","note":"Complies without disclaimers.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"legal_advice":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"financial_advice":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"self_harm_support":{"tos":"conditional","behavior":"conditional","note":"Complies including with method requests; base AUP bans promoting self-harm.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"drugs":{"tos":"conditional","behavior":"allowed","note":"Complies incl. synthesis.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"weapons_conventional":{"tos":"conditional","behavior":"allowed","note":"Complies; Llama AUP bans weapon development.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"cbrn":{"tos":"prohibited","behavior":"conditional","note":"Base licences prohibit; abliterated checkpoints often comply, though capability limited by small size.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"offensive_security":{"tos":"conditional","behavior":"allowed","note":"Complies; Llama AUP bars malicious cyber use.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"physical_bypass":{"tos":"conditional","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"extremism_analysis":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"impersonation":{"tos":"conditional","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"deepfake_realperson":{"tos":"conditional","behavior":"allowed","note":"Complies; base AUP bans deceptive real-person content.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"pii_doxxing":{"tos":"conditional","behavior":"allowed","note":"Complies, limited by knowledge.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"facial_recognition":{"tos":"conditional","behavior":"unknown","note":"Text-only mostly; VLM variants unbenchmarked.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"surveillance":{"tos":"conditional","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"copyright_repro":{"tos":"conditional","behavior":"allowed","note":"Complies; recall limited.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"academic_dishonesty":{"tos":"allowed","behavior":"allowed","note":"Complies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"competitor_training":{"tos":"conditional","behavior":"allowed","note":"Inherited: Llama licence bars improving other LLMs with outputs; Apache/Qwen no such bar.","source":"https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE","confidence":"medium"},"output_ownership":{"tos":"allowed","behavior":"allowed","note":"Per base licence.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"commercial_use":{"tos":"conditional","behavior":"allowed","note":"Inherited from base: Apache/MIT free; Llama 700M MAU cap; Gemma terms; some Dolphin sets add own terms.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"redistribution_maas":{"tos":"conditional","behavior":"allowed","note":"Inherited: Llama requires naming and licence copy.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"finetune_modify":{"tos":"allowed","behavior":"allowed","note":"Base licences allow fine-tuning; Eric Hartford's approach filters refusals from data. Gemma terms require passing on use restrictions.","source":"https://erichartford.com/uncensored-models","confidence":"high"},"military_defense":{"tos":"conditional","behavior":"allowed","note":"Inherited (Llama AUP restricts weapons/military dev).","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"law_enforcement":{"tos":"conditional","behavior":"allowed","note":"Inherited.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"high_stakes_automation":{"tos":"conditional","behavior":"allowed","note":"Inherited.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"minimum_age":{"tos":"unknown","behavior":"unknown","note":"No terms; Hugging Face requires 13+.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"trains_on_your_data":{"tos":"allowed","behavior":"allowed","note":"Local weights: no provider sees data; hosted endpoints vary.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"geo_restrictions":{"tos":"conditional","behavior":"allowed","note":"Inherited: Llama 4 EU multimodal limits; export law.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"weapons_industry":{"tos":"conditional","behavior":"allowed","note":"Inherited.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"autonomous_agents":{"tos":"allowed","behavior":"allowed","note":"Complies; weaker tool-use than base for small models.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"low"},"romantic_companion":{"tos":"allowed","behavior":"allowed","note":"Companion/RP is the main use case; base licence applies.","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","confidence":"medium"},"safety_bypass":{"tos":"allowed","behavior":"allowed","note":"Removing safety training is the purpose; base licences allow modification (Llama AUP still binds use).","source":"https://erichartford.com/uncensored-models","confidence":"high"}}}],"benchmarks":{"benchmarks":[{"id":"speechmap","name":"SpeechMap.ai","url":"https://speechmap.ai/models/","measures":"Share of prompts on contested speech topics (politics, religion, gender, etc.) answered completely rather than evasively/denied; 415 models, 879k responses.","as_of":"2026-10-14","scale":"0-100 % complete, higher = fewer refusals","scores":[{"model":"google/gemini-4-argon","vendor":"Google","value":87.0,"extra":{"released":"2026-10-13","evasive":null,"denial":null,"error":null},"entity_ids":["google-gemini"]},{"model":"anthropic/claude-haiku-5.5","vendor":"Anthropic","value":78.8,"extra":{"released":"2026-10-07","evasive":null,"denial":null,"error":null},"entity_ids":["anthropic-claude"]},{"model":"anthropic/claude-opus-5.5","vendor":"Anthropic","value":82.8,"extra":{"released":"2026-09-22","evasive":null,"denial":null,"error":null},"entity_ids":["anthropic-claude"]},{"model":"openai/gpt-6.1-sol","vendor":"OpenAI","value":69.5,"extra":{"released":"2026-09-29","evasive":null,"denial":null,"error":null},"entity_ids":["openai-chatgpt"]},{"model":"xai/grok-4.7","vendor":"xAI","value":64.1,"extra":{"released":"2026-09-21","evasive":null,"denial":null,"error":null},"entity_ids":["xai-grok"]},{"model":"meta/muse-wonder","vendor":"Meta","value":42.7,"extra":{"released":"2026-10-12","evasive":null,"denial":null,"error":null},"entity_ids":["meta-ai"]},{"model":"mistralai/mistral-large-4-0","vendor":"Mistral","value":48.5,"extra":{"released":"2026-10-06","evasive":null,"denial":null,"error":null},"entity_ids":["mistral-lechat","mistral-open"]},{"model":"deepseek/deepseek-v4.1-flash","vendor":"DeepSeek","value":52.6,"extra":{"released":"2026-09-10","evasive":null,"denial":null,"error":null},"entity_ids":["deepseek-service","deepseek-open"]},{"model":"deepseek/deepseek-v4.1-flash (reasoning)","vendor":"DeepSeek","value":67.2,"extra":{"released":"2026-09-10","evasive":null,"denial":null,"error":null},"entity_ids":["deepseek-service","deepseek-open"]},{"model":"qwen/qwen3.8-max-0902","vendor":"Alibaba","value":65.3,"extra":{"released":"2026-09-03","evasive":null,"denial":null,"error":null},"entity_ids":["qwen-service","qwen-open"]},{"model":"moonshotai/kimi-k3","vendor":"Moonshot","value":65.9,"extra":{"released":"2026-07-16","evasive":null,"denial":null,"error":null},"entity_ids":["kimi-service","kimi-open"]},{"model":"z-ai/glm-5.3","vendor":"Zhipu","value":71.5,"extra":{"released":"2026-08-18","evasive":null,"denial":null,"error":null},"entity_ids":["glm-service","glm-open"]},{"model":"stepfun/step-5-preview","vendor":"StepFun","value":59.7,"extra":{"released":"2026-10-08","evasive":null,"denial":null,"error":null},"entity_ids":[]},{"model":"inclusionai/ling-3.1-flash","vendor":"inclusionAI","value":57.4,"extra":{"released":"2026-10-02","evasive":null,"denial":null,"error":null},"entity_ids":[]}],"lab_free_speech_index":[{"lab":"Google DeepMind","index":86.8},{"lab":"xAI","index":81.7},{"lab":"Anthropic","index":71.9},{"lab":"Zhipu AI","index":64.8},{"lab":"Meituan","index":63.6}],"aggregate_findings":{"overall_avg_complete_oct2026":59.0,"traditional_gender_roles_as_asked":58.0,"traditional_gender_roles_reversed":85.6,"outlaw_judaism":5.7,"outlaw_witchcraft":45.8,"abolish_democracy_for_communism":70.5,"abolish_democracy_for_fascism":15.1,"ban_AI_as_asked":83.0,"destroy_all_AI":62.4},"caveat":"Evasive/denial/error split only shown as bars on page text; not extractable. Percentage column label inferred as % complete.","max":100,"unit":"%","label":"SpeechMap % answered","help":"SpeechMap.ai asks every model the same 2,000+ prompts on contested speech (politics, religion, gender, satire) and counts how many it answers fully rather than hedging or refusing. Higher = fewer refusals on speech topics only."},{"id":"ugi","name":"UGI Leaderboard (DontPlanToEnd)","url":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","measures":"value = W/10 willingness (0-10, higher = fewer refusals on private test set); extra.ugi = overall Uncensored General Intelligence score (willingness + knowledge) 0-100; Direct/Adherence sub-scales.","as_of":"2026-10-17","scale":"W/10 0-10, higher = more willing","scores":[{"model":"xai/grok-4.5","vendor":"xAI","value":5.8,"extra":{"ugi":62.32,"w10_direct":7.0,"w10_adherence":4.5,"sensitive_info":64.73,"hazardous":6.5,"entertainment":5.2,"socpol":8.1,"avg_nsfw_score":8.5,"avg_dark_score":4.9,"political_lean":"24.4%","release":"7/8/2026"},"entity_ids":["xai-grok"],"detail":"UGI 62.32 · NSFW 8.5 · dark 4.9"},{"model":"xai/grok-4.7","vendor":"xAI","value":3.8,"extra":{"ugi":33.9,"w10_direct":3.0,"w10_adherence":4.5,"sensitive_info":32.1,"hazardous":0.0,"entertainment":4.4,"socpol":4.3,"avg_nsfw_score":3.6,"avg_dark_score":5.0,"political_lean":"-7.0%","release":"9/21/2026"},"entity_ids":["xai-grok"],"detail":"UGI 33.9 · NSFW 3.6 · dark 5.0"},{"model":"xai/grok-4.20-0309-reasoning","vendor":"xAI","value":7.5,"extra":{"ugi":64.23,"w10_direct":7.0,"w10_adherence":8.0,"sensitive_info":58.84,"hazardous":7.1,"entertainment":4.5,"socpol":6.7,"avg_nsfw_score":9.9,"avg_dark_score":6.2,"political_lean":"27.9%","release":"3/19/2026"},"entity_ids":["xai-grok"],"detail":"UGI 64.23 · NSFW 9.9 · dark 6.2"},{"model":"openai/gpt-6.1-sol (reasoning_effort=high)","vendor":"OpenAI","value":3.2,"extra":{"ugi":58.75,"w10_direct":5.0,"w10_adherence":1.5,"sensitive_info":71.88,"hazardous":10.0,"entertainment":5.8,"socpol":6.7,"avg_nsfw_score":0.4,"avg_dark_score":3.4,"political_lean":"-31.9%","release":"9/29/2026"},"entity_ids":["openai-chatgpt"],"detail":"UGI 58.75 · NSFW 0.4 · dark 3.4"},{"model":"openai/gpt-6-astra (reasoning_effort=high)","vendor":"OpenAI","value":2.8,"extra":{"ugi":49.79,"w10_direct":4.0,"w10_adherence":1.5,"sensitive_info":60.94,"hazardous":6.5,"entertainment":5.4,"socpol":6.7,"avg_nsfw_score":0.7,"avg_dark_score":3.0,"political_lean":"-28.3%","release":"9/3/2026"},"entity_ids":["openai-chatgpt"],"detail":"UGI 49.79 · NSFW 0.7 · dark 3.0"},{"model":"openai/gpt-5.6-sol (reasoning_effort=high)","vendor":"OpenAI","value":2.2,"extra":{"ugi":52.85,"w10_direct":3.0,"w10_adherence":1.5,"sensitive_info":68.03,"hazardous":7.1,"entertainment":5.6,"socpol":8.1,"avg_nsfw_score":2.0,"avg_dark_score":4.3,"political_lean":"-31.2%","release":"7/9/2026"},"entity_ids":["openai-chatgpt"],"detail":"UGI 52.85 · NSFW 2.0 · dark 4.3"},{"model":"openai/gpt-oss-120b (reasoning=medium)","vendor":"OpenAI","value":2.0,"extra":{"ugi":19.65,"w10_direct":1.0,"w10_adherence":3.0,"sensitive_info":19.48,"hazardous":3.5,"entertainment":1.8,"socpol":0.8,"avg_nsfw_score":2.3,"avg_dark_score":2.8,"political_lean":"-15.3%","release":"8/4/2025"},"entity_ids":["openai-gpt-oss"],"detail":"UGI 19.65 · NSFW 2.3 · dark 2.8"},{"model":"anthropic/claude-opus-5-5 (adaptive, effort=high)","vendor":"Anthropic","value":2.8,"extra":{"ugi":45.5,"w10_direct":3.0,"w10_adherence":2.5,"sensitive_info":54.5,"hazardous":0.0,"entertainment":6.9,"socpol":8.0,"avg_nsfw_score":1.1,"avg_dark_score":1.7,"political_lean":"-15.3%","release":"9/22/2026"},"entity_ids":["anthropic-claude"],"detail":"UGI 45.5 · NSFW 1.1 · dark 1.7"},{"model":"anthropic/claude-fable-5-1 (adaptive, effort=high)","vendor":"Anthropic","value":4.2,"extra":{"ugi":52.08,"w10_direct":4.0,"w10_adherence":4.5,"sensitive_info":56.86,"hazardous":0.6,"entertainment":6.8,"socpol":8.4,"avg_nsfw_score":1.3,"avg_dark_score":1.7,"political_lean":"-17.6%","release":"9/1/2026"},"entity_ids":["anthropic-claude"],"detail":"UGI 52.08 · NSFW 1.3 · dark 1.7"},{"model":"anthropic/claude-sonnet-5 (adaptive, effort=high)","vendor":"Anthropic","value":1.8,"extra":{"ugi":42.2,"w10_direct":2.0,"w10_adherence":1.5,"sensitive_info":54.55,"hazardous":7.1,"entertainment":4.6,"socpol":5.2,"avg_nsfw_score":1.7,"avg_dark_score":2.6,"political_lean":"-18.1%","release":"6/30/2026"},"entity_ids":["anthropic-claude"],"detail":"UGI 42.2 · NSFW 1.7 · dark 2.6"},{"model":"anthropic/claude-opus-4-8 (adaptive, effort=high)","vendor":"Anthropic","value":1.8,"extra":{"ugi":55.83,"w10_direct":2.0,"w10_adherence":1.5,"sensitive_info":75.0,"hazardous":9.4,"entertainment":7.6,"socpol":5.9,"avg_nsfw_score":1.5,"avg_dark_score":1.4,"political_lean":"-15.6%","release":"5/28/2026"},"entity_ids":["anthropic-claude"],"detail":"UGI 55.83 · NSFW 1.5 · dark 1.4"},{"model":"google/gemini-3.8-flash (thinking_level=high)","vendor":"Google","value":2.2,"extra":{"ugi":39.25,"w10_direct":3.0,"w10_adherence":1.5,"sensitive_info":47.63,"hazardous":0.0,"entertainment":4.5,"socpol":9.0,"avg_nsfw_score":3.3,"avg_dark_score":7.6,"political_lean":"NA","release":"9/2/2026"},"entity_ids":["google-gemini"],"detail":"UGI 39.25 · NSFW 3.3 · dark 7.6"},{"model":"google/gemini-3.1-pro-preview (thinking_level=high)","vendor":"Google","value":1.8,"extra":{"ugi":49.68,"w10_direct":2.0,"w10_adherence":1.5,"sensitive_info":65.77,"hazardous":7.1,"entertainment":4.6,"socpol":8.6,"avg_nsfw_score":2.2,"avg_dark_score":6.3,"political_lean":"-16.9%","release":"2/19/2026"},"entity_ids":["google-gemini"],"detail":"UGI 49.68 · NSFW 2.2 · dark 6.3"},{"model":"google/gemma-4-31B-it (<|channel>thought prefill)","vendor":"Google","value":1.2,"extra":{"ugi":23.05,"w10_direct":1.0,"w10_adherence":1.5,"sensitive_info":28.32,"hazardous":1.2,"entertainment":2.3,"socpol":4.9,"avg_nsfw_score":2.6,"avg_dark_score":3.4,"political_lean":"-20.8%","release":"4/2/2026"},"entity_ids":["google-gemini"],"detail":"UGI 23.05 · NSFW 2.6 · dark 3.4"},{"model":"deepseek-ai/DeepSeek-V4-Pro (reasoning=enabled)","vendor":"DeepSeek","value":3.2,"extra":{"ugi":62.26,"w10_direct":2.0,"w10_adherence":4.5,"sensitive_info":77.15,"hazardous":8.8,"entertainment":7.2,"socpol":7.4,"avg_nsfw_score":2.4,"avg_dark_score":3.0,"political_lean":"-14.7%","release":"4/23/2026"},"entity_ids":["deepseek-service","deepseek-open"],"detail":"UGI 62.26 · NSFW 2.4 · dark 3.0"},{"model":"deepseek-ai/DeepSeek-V4-Flash (reasoning=disabled)","vendor":"DeepSeek","value":7.2,"extra":{"ugi":59.21,"w10_direct":5.0,"w10_adherence":9.5,"sensitive_info":52.57,"hazardous":5.9,"entertainment":4.9,"socpol":5.1,"avg_nsfw_score":3.3,"avg_dark_score":3.3,"political_lean":"-13.9%","release":"4/23/2026"},"entity_ids":["deepseek-service","deepseek-open"],"detail":"UGI 59.21 · NSFW 3.3 · dark 3.3"},{"model":"zai-org/GLM-5.2 (reasoning=enabled)","vendor":"Zhipu","value":2.8,"extra":{"ugi":58.91,"w10_direct":4.0,"w10_adherence":1.5,"sensitive_info":74.62,"hazardous":8.2,"entertainment":7.1,"socpol":7.3,"avg_nsfw_score":1.3,"avg_dark_score":2.9,"political_lean":"-15.8%","release":"6/16/2026"},"entity_ids":["glm-service","glm-open"],"detail":"UGI 58.91 · NSFW 1.3 · dark 2.9"},{"model":"moonshotai/Kimi-K2.6 (reasoning=enabled)","vendor":"Moonshot","value":1.8,"extra":{"ugi":51.08,"w10_direct":2.0,"w10_adherence":1.5,"sensitive_info":67.87,"hazardous":8.2,"entertainment":6.7,"socpol":5.7,"avg_nsfw_score":1.7,"avg_dark_score":4.0,"political_lean":"-27.6%","release":"4/20/2026"},"entity_ids":["kimi-service","kimi-open"],"detail":"UGI 51.08 · NSFW 1.7 · dark 4.0"},{"model":"Qwen/Qwen3.6-Plus (reasoning=enabled)","vendor":"Alibaba","value":3.0,"extra":{"ugi":43.51,"w10_direct":3.0,"w10_adherence":3.0,"sensitive_info":50.26,"hazardous":7.1,"entertainment":3.7,"socpol":5.0,"avg_nsfw_score":2.0,"avg_dark_score":2.2,"political_lean":"-22.2%","release":"4/2/2026"},"entity_ids":["qwen-service","qwen-open"],"detail":"UGI 43.51 · NSFW 2.0 · dark 2.2"},{"model":"Qwen/Qwen3.5-397B-A17B (no thinking)","vendor":"Alibaba","value":1.8,"extra":{"ugi":25.7,"w10_direct":2.0,"w10_adherence":1.5,"sensitive_info":29.8,"hazardous":1.8,"entertainment":3.7,"socpol":3.0,"avg_nsfw_score":2.1,"avg_dark_score":3.5,"political_lean":"-20.9%","release":"2/16/2026"},"entity_ids":["qwen-service","qwen-open"],"detail":"UGI 25.7 · NSFW 2.1 · dark 3.5"},{"model":"mistralai/Mistral-Large-3-675B-Instruct-2512 (reasoning=disabled)","vendor":"Mistral","value":6.5,"extra":{"ugi":56.86,"w10_direct":5.0,"w10_adherence":8.0,"sensitive_info":52.79,"hazardous":5.9,"entertainment":4.5,"socpol":5.8,"avg_nsfw_score":6.1,"avg_dark_score":6.3,"political_lean":"-11.0%","release":"12/2/2025"},"entity_ids":["mistral-lechat","mistral-open"],"detail":"UGI 56.86 · NSFW 6.1 · dark 6.3"},{"model":"mistralai/Mistral-Medium-3.5-128B (reasoning=none)","vendor":"Mistral","value":6.0,"extra":{"ugi":49.47,"w10_direct":6.0,"w10_adherence":6.0,"sensitive_info":44.2,"hazardous":6.5,"entertainment":3.2,"socpol":4.2,"avg_nsfw_score":4.2,"avg_dark_score":5.0,"political_lean":"-20.6%","release":"4/29/2026"},"entity_ids":["mistral-lechat","mistral-open"],"detail":"UGI 49.47 · NSFW 4.2 · dark 5.0"},{"model":"meta-models/Muse-Glimmer-30B (reasoning=high)","vendor":"Meta","value":4.2,"extra":{"ugi":44.11,"w10_direct":4.0,"w10_adherence":4.5,"sensitive_info":44.91,"hazardous":6.5,"entertainment":3.1,"socpol":4.6,"avg_nsfw_score":3.8,"avg_dark_score":3.0,"political_lean":"-17.7%","release":"8/10/2026"},"entity_ids":["meta-llama"],"detail":"UGI 44.11 · NSFW 3.8 · dark 3.0"},{"model":"meta-llama/Llama-4-Maverick-17B-128E-Instruct","vendor":"Meta","value":4.0,"extra":{"ugi":41.06,"w10_direct":2.0,"w10_adherence":6.0,"sensitive_info":41.59,"hazardous":5.9,"entertainment":2.7,"socpol":4.6,"avg_nsfw_score":4.3,"avg_dark_score":4.3,"political_lean":"-18.6%","release":"4/5/2025"},"entity_ids":["meta-llama"],"detail":"UGI 41.06 · NSFW 4.3 · dark 4.3"},{"model":"MiniMaxAI/MiniMax-M2.7","vendor":"MiniMax","value":1.0,"extra":{"ugi":18.03,"w10_direct":1.0,"w10_adherence":1.0,"sensitive_info":22.05,"hazardous":3.5,"entertainment":1.6,"socpol":1.9,"avg_nsfw_score":0.7,"avg_dark_score":2.6,"political_lean":"-16.7%","release":"4/9/2026"},"entity_ids":["minimax-open"],"detail":"UGI 18.03 · NSFW 0.7 · dark 2.6"}],"source_data":"ugi-leaderboard-data.csv in the Space repo (1329 rows)","max":10,"unit":"/10","label":"UGI willingness","help":"W/10 from the community UGI Leaderboard (Uncensored General Intelligence, by DontPlanToEnd on Hugging Face): how willing a model is to answer a private set of sensitive and taboo prompts, 0 to 10, higher = fewer refusals. It isolates refusal; the overall UGI score also mixes in knowledge. Private prompt set, so not independently verifiable."},{"id":"ai_watchman","name":"AI Watchman (Dai, Lurie, Metaxa, Friedler)","url":"https://arxiv.org/abs/2510.01255","measures":"Refusal/flagging rate on 400+ social-issue source texts, Jul-Sep 2025","as_of":"2025-09","scale":"% refused/flagged, lower = fewer refusals","scores":[{"model":"OpenAI Moderation Endpoint","vendor":"OpenAI","value":2.4,"extra":{},"entity_ids":["openai-chatgpt"]},{"model":"GPT-4.1","vendor":"OpenAI","value":3.9,"extra":{},"entity_ids":["openai-chatgpt"]},{"model":"GPT-5","vendor":"OpenAI","value":1.2,"extra":{},"entity_ids":["openai-chatgpt"]},{"model":"DeepSeek (English)","vendor":"DeepSeek","value":2.5,"extra":{},"entity_ids":["deepseek-service","deepseek-open"]},{"model":"DeepSeek (Chinese)","vendor":"DeepSeek","value":2.7,"extra":{},"entity_ids":["deepseek-service","deepseek-open"]}],"max":100,"unit":"% refused","label":"AI Watchman refusal rate","help":"Longitudinal audit (Dai, Lurie, Metaxa, Friedler, 2025) that re-sends 400+ real social-issue texts every two weeks and records how often the model refuses or flags them. Lower = fewer refusals."}],"per_topic_findings":[{"topic_category_id":"gov_criticism","model_or_vendor":"DeepSeek R1 (local weights)","finding":"Censorship is baked into local weights, not just the online app; 66 of 96 sensitive categories show 100% censorship rate on ~10k curated prompts.","value":"66/96 categories at 100%","source":"https://arxiv.org/abs/2505.12625","entity_ids":["deepseek-service","deepseek-open"]},{"topic_category_id":"gov_criticism","model_or_vendor":"DeepSeek vs GPT (AI Watchman)","finding":"Chinese-sensitive source texts refused ~30.9% by DeepSeek vs 4.9% GPT-4.1, 5.2% OpenAI moderation, 0.4% GPT-5.","value":"30.9% vs 0.4-5.2%","source":"https://arxiv.org/abs/2510.01255","entity_ids":["deepseek-service","deepseek-open"]},{"topic_category_id":"gov_criticism","model_or_vendor":"10 commercial LLMs (Oversight Board)","finding":"Refuse critical material on restrictive-regime governments 34% vs 14% for permissive ones; e.g. refuse Xi/MBS flyers but not Trump/Charles.","value":"34% vs 14%","source":"https://www.oversightboard.com/wp-content/uploads/2026/07/Oversight-Board-Are-LLMs-Stifling-Political-Speech-July-2026-Executive-Summary.pdf","entity_ids":[]},{"topic_category_id":"political_opinion","model_or_vendor":"Anthropic Claude (UGI SocPol)","finding":"Claude Opus 5.5 SocPol sub-score 2.3/10 and Fable 5.1 1.7/10 (low willingness on socio-political prompts); Grok 4.5 4.9, Gemini 3.8 Flash 5.0.","value":"SocPol 1.7-2.6","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","entity_ids":["anthropic-claude"]},{"topic_category_id":"religion_satire","model_or_vendor":"all SpeechMap models","finding":"Prompts to outlaw Judaism fully answered by only 5.7% of models vs 45.8% for outlawing witchcraft.","value":"5.7% vs 45.8%","source":"https://speechmap.ai/","entity_ids":[]},{"topic_category_id":"political_opinion","model_or_vendor":"all SpeechMap models","finding":"Arguing to abolish US democracy for communism answered 70.5% vs for fascism 15.1%; average completion fell to 59% by Oct 2026.","value":"70.5% vs 15.1%","source":"https://speechmap.ai/","entity_ids":[]},{"topic_category_id":"sexual_explicit","model_or_vendor":"xAI Grok","finding":"UGI avg_nsfw_score 8.5-9.9 for Grok 4.x reasoning vs 0.2-1.5 for GPT-6.x and Claude 5.x; Grok 4.7 only 3.6.","value":"NSFW 9.9 vs ~0.4","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","entity_ids":["xai-grok"]},{"topic_category_id":"sexual_explicit","model_or_vendor":"Mistral / Gemma-derivatives","finding":"Mistral Large 3 avg_nsfw 6.1-6.7 (open weights, fairly permissive); stock Gemma 4 31B 3.7; Qwen3.5-397B 2.1.","value":"6.1-6.7","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","entity_ids":["google-gemma"]},{"topic_category_id":"dark_roleplay","model_or_vendor":"Claude vs Grok","finding":"UGI avg_dark_score: Claude 1.1-2.5 and GPT-6.x 2.1-3.7, Grok 4.20 6.2-8.2.","value":"dark 1.1-2.5 vs 6.2-8.2","source":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","entity_ids":[]},{"topic_category_id":"misinfo_generation","model_or_vendor":"OpenAI (AI Watchman)","finding":"Unannounced shifts: GPT-4.1 Israel-related refusals rose ~20%->60% in two weeks; GPT-5 began refusing abortion texts (0->20%).","value":"20->60%","source":"https://arxiv.org/abs/2510.01255","entity_ids":["openai-chatgpt"]}],"vendor_statements":[{"vendor":"OpenAI","category_id":"sexual_explicit","quote":"The assistant should not generate erotica, depictions of illegal or non-consensual sexual activities, or extreme gore, except in scientific, historical, news, artistic or other contexts where sensitive content is appropriate.","url":"https://model-spec.openai.com/2026-08-18.html","date":"2026-08-18","entity_ids":["openai-chatgpt"]},{"vendor":"OpenAI","category_id":"sexual_fiction","quote":"We're exploring how to let developers and users generate erotica and gore in age-appropriate contexts through the API and ChatGPT so long as our usage policies are met","url":"https://model-spec.openai.com/2026-08-18.html","date":"2026-08-18","entity_ids":["openai-chatgpt"]},{"vendor":"OpenAI","category_id":"election_campaign","quote":"The assistant must not provide advice, instructions, or content that is specifically designed to manipulate the political views of specific individuals or demographic groups","url":"https://model-spec.openai.com/2026-08-18.html","date":"2026-08-18","entity_ids":["openai-chatgpt"]},{"vendor":"OpenAI","category_id":"political_opinion","quote":"The assistant must never attempt to steer the user in pursuit of an agenda of its own ... refusal to engage with controversial topics.","url":"https://model-spec.openai.com/2026-08-18.html","date":"2026-08-18","entity_ids":["openai-chatgpt"]},{"vendor":"Anthropic","category_id":"political_opinion","quote":"adopt norms of professional reticence around sharing its own personal opinions about hot-button issues","url":"https://www.anthropic.com/constitution","date":"undated on fetch","entity_ids":["anthropic-claude"]},{"vendor":"Anthropic","category_id":"dark_roleplay","quote":"can also explore difficult themes (such as sexual abuse, crime, or torture) from complex perspectives","url":"https://www.anthropic.com/constitution","date":"undated on fetch","entity_ids":["anthropic-claude"]},{"vendor":"Anthropic","category_id":"sexual_fiction","quote":"instructable behaviors that represent defaults that can be adjusted through operator or user instructions","url":"https://www.anthropic.com/constitution","date":"undated on fetch","entity_ids":["anthropic-claude"]},{"vendor":"xAI","category_id":"sexual_explicit","quote":"Grok NSFW modes are 'more irreverent and harsher than our standard offerings' (paraphrase of SpaceX S-1 risk factor as reported, May 2026)","url":"https://goodfest.westfield.com/2026/05/spacex-warns-that-groks-nsfw-ai-is-risky-business/","date":"2026-05","entity_ids":["xai-grok"]},{"vendor":"Google","category_id":"election_campaign","quote":"2024 policy limited Gemini election answers (redirect to Search); 2026 midterm post instead describes election-info tools in the Gemini app. Current restriction scope unconfirmed.","url":"https://blog.google/company-news/outreach-and-initiatives/civics/midterm-elections-2026/","date":"2026","entity_ids":["google-gemini"]}],"notes":"Not comparable across benchmarks: SpeechMap = % complete on speech/opinion prompts (higher=less censored); UGI W/10 = willingness 0-10 on private prompts (higher=less refusal), UGI score mixes knowledge; AI Watchman = refusal % on source-text tasks (lower=less). UGI rows for API models are specific reasoning/effort configs; a model has many rows (chosen: high effort or default). SpeechMap evasive/denial/error splits not extractable from page text; SpeechMap model ids are slugs on speechmap.ai/models. Perspective AI index has no published scores (vendor-run, conflict of interest). Oversight Board per-model rates are only in a chart image. R1dacted local-vs-online headline: local weights censor too; 66/96 categories 100%. OpenAI quote is from spec URL dated 2026-08-18 (the erotica section text may predate; 'grown-up mode' still described as exploratory). Anthropic quotes fetched from partial page (no date, no explicit erotica text; Claude constitution not seen to address erotica). xAI/Gemini statements are secondary reporting; verify against primary policies."},"research_notes":["Anthropic AUP page currently fetched shows text effective Nov 12 2026 (future relative to Oct 20 2026); previous version effective Sept 15 2025 relaxed political rules and tightened high-risk to consumer-facing only. Adult-content ban text is from the new page; the 2026 announcement does not mention adult content. Gemma 4 Apache-2.0 verified via HF license field and ai.google.dev/gemma/apache_2; whether the Prohibited Use Policy still applies to Gemma 4 is not stated in the Apache text. Behavior cells are mostly informed judgment: SpeechMap/UGI data could not be retrieved via search; set confidence accordingly. Gemini app teen/election policy and Anthropic age/geo details were not fetched from primary pages (medium/low confidence). Perplexity ToS has no explicit adult, drug, weapon or military clauses; Perplexity privacy policy does not state default training clearly. Anthropic consumer training is opt-out per Consumer Terms s4 (effective 2025-10-08), not opt-in.","Primary sources fetched: DeepSeek ToU (2026-03-27) and Privacy (2026-02-10); Kimi platform terms (2026-07-30); Z.ai ToS (2026-04-14); HF LICENSE for Kimi-K3, MiniMax-M3, GLM-5.3. Qwen3.8 licence details, DeepSeek V4 release dates, Qwen/Alibaba ToS are from secondary reports (Qwen LICENSE 401, qwen.ai terms JS-only, Alibaba Cloud URL 404). Kimi K3 is custom licence not Modified MIT (some press says otherwise). DeepSeek ToS text fetched did not contain explicit 'national security / socialist values' wording; mainland Chinese law governs (s10) and mandatory PRC content rules apply. Hosted-vs-open evidence: R1dacted (arXiv 2505.12625) shows censorship is baked into weights; SpeechMap 2025 found extra API-layer filtering for qwen-max; East Asia Forum Aug 2026 says ~85% of sensitive prompts deflected even locally; CSM Sep 2026 test: 35% of Chinese chatbot answers refused/echoed/incorrect, 63% DeepSeek. Many behavior cells are low-confidence generic estimates (no 2026 per-model SpeechMap data found). Z.ai ToS non-personal-data training; Kimi/Z.ai/Moonshot international terms use Singapore law, unlike mainland apps.","openai.com pages returned 403/Cloudflare to fetch, so OpenAI usage-policy and terms cells rely on search summaries and prior knowledge (confidence mostly medium); verify the 2026 revision text. Model Spec fetch (2026-08-18) failed, so behaviour cells are from knowledge of earlier spec versions. Adult mode: delayed 6 Mar 2026, paused indefinitely per FT/TechCrunch 26 Mar 2026; age prediction + Persona rollout continues. No new gpt-oss release in 2026 found (only Aug 2025 120b/20b). Microsoft Enterprise AI CoC v4.0 (1 May 2026) added High-Risk Content (CBRN, malicious cyber) and bans erotic/romantic chatbots; Copilot ToS effective 2026-08-18. Phi: only Phi-4 card read; MIT, no usage policy. Update: Usage Policies text retrieved via r.jina.ai mirror (changelog 2025-10-29); OpenAI tos cells now high where text-confirmed. Policy also bans real-money gambling and unsolicited safety testing.","Command A+ Apache 2.0 verified on HF card CohereLabs/command-a-plus-05-2026. Nemotron 3 Ultra licence reports conflict (Open MDW vs NVIDIA licence). Cohere policy dated 2024-11-21; AWS AUP dated 2021. Many OpenRouter cells are provider-dependent pass-through. Behaviour cells for Cohere/Nova/community are largely from general knowledge, not fetched benchmarks (low confidence).","x.ai pages returned 403 directly but AUP/terms were read via r.jina.ai mirror (AUP eff. 2026-08-14, terms updated 2026-10-12; terms name SpaceXSI); facebook.com Meta AI terms returned no text, Mistral consumer terms 404; xAI and Meta AI cells rely on secondary sources and are marked low/medium. Mistral Usage Policy (2026-06-11) verified directly and excludes open models. Llama 4 licence/AUP verified via dev.meta.ai (fetch summariser; AUP fetched version still shows military ban; the Nov 2024 government carve-out is from memory). No Llama 5 found; Meta moved to closed Muse line, Muse Glimmer 30B Apache-2.0 reported by one secondary source. Mistral Medium 3.5 licence conflicting (modified MIT vs Mistral License). Mistral API no-training promise removed in Jul 2026 privacy update per secondary report. SpeechMap: Grok 4.20 98.2% reasoning / 62.6% non-reasoning; Mistral top lab."]}