Amazon Bedrock Guardrails launches InvokeGuardrailChecks API for agentic AI

AWS announced a new resourceless API for Amazon Bedrock Guardrails — InvokeGuardrailChecks — that lets developers apply individual safety checks at any point in an agentic AI application without first provisioning a guardrail resource. The 'resourceless' design removes setup friction and gives per-request, per-step control over exactly which safeguards run.

The key value is granularity for agent loops: rather than applying a monolithic guardrail to an entire interaction, developers can insert specific checks at each step of an agent's reasoning-and-action loop and get back numeric severity and confidence scores for custom handling logic. That lets builders, for example, run a strict check before a tool call that touches external systems and a lighter one on intermediate reasoning.

The launch underscores how 'agent safety' has become a first-class enterprise concern as autonomous agents proliferate — a theme echoing AWS's ServiceNow governance collaboration for 'runaway AI agents' and the broader industry reaction to the Anthropic jailbreak this week. Returning numeric scores rather than binary pass/fail signals AWS expects sophisticated, programmable safety pipelines.

Competitively, this strengthens Bedrock against Azure and Google's agent platforms by making safety composable at the API level. Watch for pricing per check, latency overhead added to each agent step, and how the severity/confidence scoring compares to standalone moderation services.