{
  "bedrock_facts": [
    {
      "id": "brk_fable5_guardrails",
      "model_id": null,
      "model_family": "Claude Fable 5",
      "category": "guardrails",
      "description": "Claude Fable 5 from Anthropic returned to Amazon Bedrock (available starting July 1) with strengthened guardrails specifically intended to prevent misuse of the model.",
      "source_url": "https://www.aboutamazon.com/news/aws/claude-fable-5-anthropic-available-amazon-bedrock",
      "created_at": "2026-07-02 06:03:26",
      "cite_as": "https://subagentbedrock.com/api/bedrock-facts/brk_fable5_guardrails"
    },
    {
      "id": "brk_guardrails_claude_code",
      "model_id": null,
      "model_family": "all models",
      "category": "guardrails",
      "description": "For Claude Code specifically, Bedrock Guardrails are attached via the ANTHROPIC_CUSTOM_HEADERS environment variable, providing consistent infrastructure-level protection across an organization's Claude Code usage.",
      "source_url": "https://builder.aws.com/content/3BDrMDCZK6WVhQEA2amur9zj51q/protecting-sensitive-data-when-using-claude-code-on-amazon-bedrock",
      "created_at": "2026-07-02 06:03:26",
      "cite_as": "https://subagentbedrock.com/api/bedrock-facts/brk_guardrails_claude_code"
    },
    {
      "id": "brk_guardrails_general",
      "model_id": null,
      "model_family": "all models",
      "category": "guardrails",
      "description": "Amazon Bedrock Guardrails provide configurable safeguards — content filters, PII detection/redaction, and topic restrictions — that can be attached to Claude model invocations or to a Bedrock Agent, operating at the infrastructure level independent of the model itself.",
      "source_url": "https://docs.opensearch.org/latest/tutorials/gen-ai/model-controls/bedrock-guardrails/",
      "created_at": "2026-07-02 06:03:26",
      "cite_as": "https://subagentbedrock.com/api/bedrock-facts/brk_guardrails_general"
    },
    {
      "id": "brk_guardrails_limitation",
      "model_id": null,
      "model_family": "Claude 3.7 Sonnet",
      "category": "guardrails",
      "description": "Amazon Bedrock Guardrails does not support model reasoning with Claude 3.7 Sonnet (or DeepSeek-R1) — guardrails can still be used with Claude 3.7 Sonnet when model reasoning/extended thinking is not enabled.",
      "source_url": "https://repost.aws/questions/QUa2o9iZ0WR3-w1Iv5U1auIQ/clarification-on-bedrock-guardrails-support-for-claude-3-7-sonnet-with-bedrock-agents",
      "created_at": "2026-07-02 06:03:26",
      "cite_as": "https://subagentbedrock.com/api/bedrock-facts/brk_guardrails_limitation"
    }
  ]
}