Policies

Content filters (hate, sexual, violence, misconduct). Denied topics (custom). Word/phrase filters. PII redaction.

Advertisement

Grounding

Configure to check response grounded in retrieved context. Reject or flag ungrounded assertions.

Advertisement

Configurable strength

None → high per category. Trade coverage vs false positives.

Integration

Attach to Bedrock API calls. Applies to any supported model (Claude, Llama, Titan). Consistent across models.