Skip to main content
Guardrails check the inputs going into an LLM and the outputs coming back, and can correct unsafe content before it reaches a user. Arize AX works with guardrails in two ways. You can trace the guardrails already running in your AI agent so you can see which checks ran and how each one came out, and you can attach an Amazon Bedrock guardrail to an AI provider integration.

Trace guardrail checks

When you trace an AI agent that uses guardrails, the guardrail steps appear as their own span kind, so each check is visible in the trace. Setup is covered in the Guardrails AI tracing guide.

Use an Amazon Bedrock guardrail

An AI provider integration can use an Amazon Bedrock guardrail. The integration takes a Guardrail ID or ARN and a Guardrail Version, where the version is a version number or DRAFT.