Amazon Web Services (AWS) launched two new capabilities for Guardrails for Amazon Bedrock, its service for implementing safeguards based on application requirements and responsible AI policies.

The first capability is contextual grounding checks, which can detect hallucinations in model responses based on a reference source and user query. It filters responses that are not grounded in enterprise data or irrelevant to the user's query, improving response quality for use cases like question-answering, summarization, and information extraction.

The second capability is the ApplyGuardrail API, which allows customers to evaluate input prompts and model responses against configured safeguards for any foundation model (FM), including custom and third-party FMs outside of Amazon Bedrock. This enables centralized governance across all GenAI applications.

Analyst QuickTake: AWS initially announced Guardrails in November 2023 , geared toward implementing safe AI usage in line with company policies. The new update extends this functionality to a wider customer base, including those who do not use Amazon Bedrock specifically via its stand-alone API.