In brief: The new API enables granular application of safeguards at every point in multi-turn agent loops and allows defining custom thresholds and actions (block, bypass, retry) based on numerical scores.
AWS introduces the InvokeGuardrailChecks API for Amazon Bedrock Guardrails, allowing developers to flexibly integrate security checks into every step of agent-based AI applications — without having to provision separate Guardrail resources.
AWS announces the InvokeGuardrailChecks API for Amazon Bedrock Guardrails. The API operates without prior resource provisioning and enables deployment of individual safeguards at any point in agent-driven AI workflows. The system delivers numerical ratings for each check performed and operates in detect-only mode, without automatically blocking content.
Agent-based AI systems differ significantly from classical prompt-response scenarios: they undergo multiple cycles (multi-turn workflows) with different risk profiles per step. An example customer support agent could have 10, 20, or more iterations in a session — such as a user question (risk: prompt injection), a model response (risk: harmful content), a follow-up question with sensitive data (risk: PII), and a final response. Each point requires different security checks. With classical Guardrail resources, operational overhead arises that quickly becomes unmanageable with hundreds of agents.
The new API uses a structured message schema with role-based content blocks (system, user, assistant), providing context for precise security assessments. It is resource-agnostic — no need to create Guardrail objects, manage IDs, or version numbers. Developers define their own thresholds and execute retry, block, or bypass logic as needed within their application — for example, to log results.
Thus, the solution supports the comprehensive security controls of Amazon Bedrock Guardrails — detection and filtering of unwanted content and protection of sensitive information — even for complex, iterative agent workflows, without requiring separate resources to be provisioned for each agent stage.
Source: aws.amazon.com · Published June 17, 2026
Lumi AI News — AI-assisted curation pursuant to Art. 50 EU AI Act. Paraphrase and classification by Lumi News Pipeline v1.7.1.