Use Bedrock Guardrails with the /messages and /prompt endpoints

The /messages and /prompt endpoints include built-in guardrails. These guardrails help ensure responsible and ethical use of AI.
These guardrails help identify and block content in these categories:
  • Hate
  • Insults
  • Sexual
  • Violence
  • Misconduct

To enable these guardrails, set the safetyGuardrail field to true when calling the endpoints.

Response example

This is a response example that you can get when the guardrail conditions are met:

{
  "warnings": [],
  "id": "47add181-9684-41df-9cad-02af925af8ec",
  "reasoning_content": null,
  "content": "Blocked due to policy violation.",
  "model": "CLAUDE",
  "version": "claude-3-7-sonnet-20250219-v1:0",
  "token_usage": {
    "input": 0,
    "output": 0,
    "total": 0,
    "billed": true,
    "cache_create": 0,
    "cache_read": 0
  },
  "guardrail_info": {
    "tenant_guardrails_applied": [],
    "bedrock_guardrail_applied": true,
    "type": "VIOLENCE",
    "message": "guardrail_intervened",
    "input_flagged": true,
    "output_flagged": false
  }
}