Safeguard your agentic AI applications with the Amazon Bedrock Guardrails InvokeGuardrailChecks API

This article announces the InvokeGuardrailChecks API for Amazon Bedrock Guardrails, a new resourceless API that enables targeted safety checks at specific points in agentic AI applications.

• Apply individual safeguards (content filters, prompt attack detection, sensitive information filters) at any turn in agent loops without creating guardrail resources
• Returns numeric severity and confidence scores instead of blocking content, allowing custom thresholds and application-driven decisions
• Resourceless design eliminates overhead of creating, versioning, and managing separate guardrail resources for each workflow stage
• Supports independent prompt attack detection with fine-grained categories (jailbreak, prompt injection, prompt leakage)
• Integrates with agent frameworks via lifecycle hooks for checking user input, tool output, and final responses
• Provides precise location data (character offsets) for sensitive information findings to enable client-side redaction

The InvokeGuardrailChecks API enables developers to build safer multi-turn agentic AI applications with granular, context-aware safety controls tailored to each workflow stage.