Grok API Adds Request-Level Moderation
xAI now lets API customers add a `guidelines` array to Chat Completions and Responses requests, controlling how Grok handles NSFW, violence, and profanity. The feature currently supports only `decline` behavior.
Request-level moderation gives developers more predictable control over production outputs, but it is a narrow first step rather than a fully configurable safety layer.
- –Guidelines apply only to the request that carries them, unless repeated across a conversation
- –Omitted categories retain Grok’s default moderation behavior
- –The controls can reduce reliance on external filtering for customer-facing applications
- –Multi-agent models, deferred completions, and some `stop` or `n` configurations are unsupported
- –Developers should test refusal behavior and error handling before deploying these policies broadly
DISCOVERED
1h ago
2026-10-29
PUBLISHED
3h ago
2026-10-28
RELEVANCE
AUTHOR
grok