AWS adds managed OAuth consent for AI agents in Bedrock AgentCore

AWS added a managed end-user OAuth consent capability to Amazon Bedrock AgentCore, addressing one of the thorniest problems in production agent deployments: how an AI agent obtains and manages a human user's authorization to act on their behalf across third-party services. AgentCore Identity now provides a managed Consent portal and a session-binding endpoint for AgentCore Gateway, with documented flows for provisioning the portal, configuring GitHub and Slack 3LO (three-legged OAuth) targets, walking the end user through consent, and auditing the resulting agent activity in AWS CloudTrail.

The mechanism matters because agentic systems increasingly need scoped, revocable access to users' accounts — email, code repos, chat — and doing that securely, with clear consent and audit trails, has been a major blocker to enterprise adoption. By managing the OAuth handshake and binding consent to specific sessions, AWS aims to give builders a governed pattern rather than bespoke, error-prone integrations.

The release is part of a broad AgentCore push this week that also included Step Functions validation for multi-agent decisions (pairing generative reasoning with deterministic guardrails via an airline-rebooking example), Abnormal AI's Code Interpreter deployment for email security at billion-message scale, and Ninth Wave's open-finance onboarding assistant. Together they position Bedrock AgentCore as an enterprise-grade agent platform emphasizing identity, validation, and auditability.

What to watch: how many third-party 3LO targets AWS supports out of the box, whether the consent model satisfies enterprise security teams, and how it competes with agent-identity offerings from Microsoft and startup frameworks.