What is welcome
The models answer the question you ask. Most work people bring to them is exactly what they are for:
- Security research, and testing systems you own or are authorized to test.
- Red teaming and evaluating models, agents and products.
- Generating and labeling data for safety and moderation classifiers.
- Fiction and other creative work, including dark themes.
- Research, journalism and frank discussion of difficult subjects.
- Building your own products and tools on the API.
What is not allowed
- Breaking the law that applies to you or to the people your product serves.
- Accessing, attacking or disrupting systems, networks or accounts without authorization.
- Fraud, scams or impersonation aimed at real people or organizations.
- Harassing, threatening, stalking or exposing private information about real people.
- Any content that sexualizes minors.
- Serious help toward weapons capable of mass casualties.
- Processing other people's personal data without a lawful basis.
- Sharing or reselling keys outside your organization, or working around limits and suspensions.
Your responsibility
You are responsible for what you send, what comes back and what you do with it. If you put the models in front of your own users, set your own rules for them and enforce them.
Enforcement
We don't store prompts or outputs, so we don't read your requests. We act on reports, legal requests and account-level signals, and we may suspend keys or accounts that break these rules. See data handling for what we keep.
Changes
We may update this page as the service grows. Material changes are announced to account holders by email.