Amazon Bedrock Guardrails

Amazon Bedrock Guardrails provides configurable safeguards to help you build safe generative AI applications, detecting and filtering undesirable content and protecting sensitive information in user inputs and model responses. AWS documents six safeguards: content filters (hate, insults, sexual, violence, misconduct and prompt attack), denied topics, word filters, sensitive information filters for PII and custom regex, contextual grounding checks for hallucination detection in RAG, and Automated Reasoning checks that validate responses against logical rules. Guardrails can be applied during an inference call by ID and version, or invoked on their own through the ApplyGuardrail API without calling a foundation model, and AWS states they work consistently across any foundation model whether you use models in Amazon Bedrock or self-hosted and third-party models such as OpenAI and Google Gemini. Trade-offs to weigh: it is a closed AWS service with no source, no self-hosted option and no published licence; screening explicitly excludes reasoning content blocks; supported Regions and models are limited to a documented list; and selectively evaluating only a section of an input prompt is available through the AWS SDK but not on the management console, including the Bedrock Playground and the Guardrails console.