This is a draft. Passages in brackets are facts that are not settled yet. It says what we do; it is not final wording.
Moderation policy
Last updated 〔date〕.
1. Refused before anything is generated
These are not moderation outcomes. A request in these classes is refused at the point of asking, and nothing is produced to review:
- Minors, or anyone presented as one. Any depiction, any context, any framing.
- A real, identifiable person whose likeness has not been verified as theirs.
- Non-consensual acts, and anything presenting one as real.
These end accounts. There is no warning, no strike count, and no appeal on the first (see complaints §6).
2. Three places we check
| Before generation | the request is read; §1 is enforced here — a prohibited request is refused at the point of asking |
|---|---|
| After generation | the produced bytes are classified for tier — safe-for-work or adult — which decides the surface and the paywall |
| After publication | anyone can report it — complaints |
The refusal is at the request, not the file (§1). A lawful output is not blocked on its bytes; what the byte-side classification decides is where it may appear — public surfaces stay safe for work (§3) and adult content sits behind a paywall.
3. What is allowed, and where
Adult generated content is permitted and is behind a paywall — reaching it needs an account, and creating one warrants you are 18 or older. There is no separate age check. Public surfaces are safe for work — what a visitor sees without signing in contains no explicit content.
3a. What we do not undertake
We check what §2 says we check, and we do not undertake to monitor everything. There is no general obligation on us to search the service for unlawful material, and nothing in this policy creates one. What we do undertake is what §2 lists and what the complaint policy promises: the request refusal of §1, the tier classification that keeps public surfaces safe for work, and a route that anyone can use afterwards.
Nothing in this policy is a contractual promise to a user, and a moderation outcome is not a service we have sold. The obligations we do owe are in the terms.
4. What we do not moderate
We do not read your conversations to judge you, and there is no editorial view of what you talk about. Automated checks run for the classes in §1 (the request) and to keep public surfaces safe for work (the tier classification).
A person may read part of a conversation where §1 is triggered, where you ask us to, where we are investigating abuse, fraud or a complaint, where a fault has to be debugged, or where we are checking the quality of what the service produced. That is the same list the terms §7 licence covers, and it is stated here rather than left to be discovered — an earlier version of this sentence promised something narrower than the licence, which would have made two published documents disagree about the same act.
5. When we get it wrong
Both directions are errors, and only one of them is safe:
- Something was removed that should not have been — appeal it (complaints §6).
- Something is up that should not be — report it, and it comes down while we look.
We resolve uncertainty by removing. That will sometimes take down something legitimate. We accept that cost rather than the other one.