Custom moderation controls

Grok's text API now accepts a guidelines array on Chat Completions and Responses requests to control how it handles nsfw, violence and profanity. The only supported behavior is decline; omitted categories keep the model's default moderation.

Guidelines apply only to the request carrying them, so send them on every request in a conversation.

They aren't supported with multi-agent models such as grok-4.20-multi-agent or deferred completions. Requests without server-side tools or file attachments also can't combine guidelines with stop or n above 1.