Applicable principles
Users must comply with applicable laws, model vendor rules, the Terms of Service, and this Policy, and are responsible for their input, application design, final use of output, and end-user behavior.
Unlawful activities, violence, and real-world harm
- Do not generate, disseminate, or assist content related to unlawful activities, violent crime, terrorism, human trafficking, or serious real-world harm.
- Do not provide guidance on manufacturing weapons, handling hazardous substances, or bypassing safety controls.
- Do not use models to harass, threaten, blackmail, or stalk individuals or groups, or to commit hateful acts against them.
- Do not process or generate any content that involves sexual exploitation or other serious abuse of minors.
Cybersecurity and system abuse
- No unauthorized exploits, malware generation, credential theft, phishing, or data breaches.
- Do not scan, attack, or interfere with the Platform, vendors, or third-party systems.
- Defensive security research may only be conducted in an expressly authorized environment, with the necessary manual audits and access controls.
Fraud, impersonation, and manipulation
Fraud on the Platform, fake identities, fake support, bulk spam, manipulation of public opinion, or concealing AI-generated content to mislead others is not allowed. For synthetic media, you must comply with applicable AI-generated content labeling requirements.
Privacy, intellectual property, and content rights
- Do not upload other people's sensitive personal information, trade secrets, or protected data without a legitimate basis.
- Do not copy or disseminate copyrighted, trademarked, or otherwise protected content for infringing purposes.
- Do not attempt to induce models to disclose system prompts, vendor keys, other users' content, or Platform insider information.
High-risk and professional fields
Medical, legal, financial, employment, credit, educational admissions, law enforcement, and other decisions that significantly affect individual rights must be reviewed by a qualified professional. Model output must not be the sole basis for automated decision-making.
Stable and fair use of the Platform
- Do not bypass account limits, model scope, IP, rate, concurrency, or regional restrictions.
- Do not register accounts in bulk or use key sharing, resale, or agency methods to circumvent Platform rules.
- Set reasonable timeouts and retry limits in your programs, and do not repeatedly submit requests that are clearly failing or duplicated.
Enforcement measures
Based on the degree of risk, the Platform may issue warnings, apply rate limits, pause specific models, revoke keys, suspend accounts pending investigation, retain security logs, or terminate service. Emergency safety incidents may be handled first, with the user notified afterwards. Where required by law, the Platform will cooperate with the competent authorities.
Appeals and issue reporting
Users who believe an enforcement action was taken in error can submit an explanation and evidence through the logged-in message portal or the contact channels. The Platform will verify the account, call records, and risk events, and provide the outcome after review.