Inference hooks are out of beta for Claude Enterprise organizations

Claude Enterprise’s inference hooks are out of beta. They route governed prompts through a security server run by your organization or vendor for an allow or deny verdict before the model runs.

Setup requires the organization:manage permission, held by Owner and Primary owner roles. The hooks don’t cover Claude Platform API organizations, Amazon Bedrock or Google Cloud.

The server receives transcript text, tool calls and results, and text extracted from attachments. It can’t inspect image-only content or rewrite prompts. Administrators choose whether server failures block requests or allow them through without inspection. Anthropic advises against using hooks as the only control for voice mode.