Amazon Bedrock Guardrails

Amazon Bedrock Guardrails is a managed feature of AWS's Bedrock generative-AI service that lets organizations apply configurable safeguards to model inputs and outputs. Rather than a standalone company, it is a cloud capability that evaluates both prompts and responses against a set of policies teams define. Its safeguards include content filters across categories such as hate, insults, sexual content, violence, misconduct, and prompt attacks with adjustable strength; denied-topics controls to keep applications on-topic; word and profanity filters; sensitive-information filters that detect and redact PII; contextual grounding checks that flag hallucinated or irrelevant responses against source material; and Automated Reasoning checks that validate outputs against formal logical rules to catch factual errors. Guardrails can be applied to models hosted in Bedrock as well as self-managed or third-party models via an independent ApplyGuardrail API, and can protect agents and RAG workflows. Pricing is usage-based per 1,000 text units, with AWS having cut content-filter and denied-topic costs by up to 85% in late 2024. Aimed at enterprises building on AWS, its distinguishing strengths are deep native integration with the Bedrock stack, IAM-governed configuration, and the Automated Reasoning approach to hallucination prevention.