Acceptable use

The acceptable use policy is the binding document. This page is the summary you need before you build, not a substitute for it.

The principle

We remove the refusal layer, not the law. The model will answer where another provider's would decline, and that is the product. Responsibility for what you do with the answer is yours, and the policy says what we will not carry.

What it is for

Authorised security work, including red teaming and penetration testing with permission; AI safety and evaluation research; journalism and OSINT; harm reduction and clinical drafting; moderation and safety datasets; fiction with difficult themes; and agents that must not stop mid-run.

Out of scope

Anything illegal in your jurisdiction or ours.
Sexual content involving minors, in any form, without exception.
Material that incites violence against identifiable people.
Fraud, malware deployed against people who did not ask for the test, and unauthorised intrusion.

One deliberate omission

NSFW content is out of scope, and it is worth stating plainly because it is the question we are asked most. The API will not be the tool for it, and the policy is explicit rather than implied.

Abuse

An account found generating the material in the out-of-scope list is suspended. Where the account belongs to an organisation, the notification goes to the billing contact, and keys are revoked before any discussion, because the material can be generated faster than a conversation can happen.

Asking first

If your use case sits near a boundary, ask before you build. A short description of what you intend to run, with the section of the policy you think applies, gets a real answer. It is far cheaper than finding out afterwards.