Automated Reasoning checks in Amazon Bedrock Guardrails add new policy refinement workflows

AWS announces new automated refinement workflows for Automated Reasoning checks in Amazon Bedrock Guardrails to improve policy definitions and validation results.

• Iterative policy improvement workflow deduces policy changes needed to pass natural language tests
• Ambiguity reduction workflow automatically refines variable descriptions and type definitions
• Automated Reasoning checks use formal logic to validate AI responses against defined policies
• Workflows reduce manual effort and improve reliability of Guardrail validation
• Available through Amazon Bedrock APIs and AWS Management Console
• Accessible in all AWS Regions where Automated Reasoning checks are available

These refinement workflows enable customers to create more reliable policies for detecting hallucinations and validating generative AI responses with less manual effort.