AWS Shows How to Govern AI Agent Access With Okta Cross App Access

Published November 6, 2026, AWS’s Part 1 walkthrough builds a task agent on Amazon Bedrock AgentCore that acts for signed-in users through Okta Cross App Access and AgentCore Identity. Okta Cross App Access evaluates policy whenever the agent requests an action. Okta’s documentation describes the broader flow as validating an ID-JAG and exchanging it for a short-lived, scoped access token.

In AWS’s described architecture, AgentCore runtime validates Okta’s grant, then AgentCore Identity redeems it for an access token and stores that token in its token vault. AWS reports that removing a user’s assignment led Okta to refuse new requests for that user within about one second; already-issued tokens expired within five minutes. The ID-JAG is single-use and expires in five minutes, as does the resulting Okta access token, which identifies both the user and agent and carries the user’s groups.

In testing, AgentCore Identity added about 0.6 seconds to a request that took about 30 seconds end to end. Across 10 requests, a second token-vault lookup for another MCP server returned the same token in a median of 58 milliseconds without redeeming the single-use ID-JAG again. The tested write authorization required both the user’s group membership and the xaa:write scope.

AWS estimates that following the walkthrough takes about an hour and says AgentCore charges for usage. Prerequisites include an AWS account with AgentCore access, an Anthropic Claude model in US West (Oregon), and an Okta org with Cross App Access and API Access Management. The post says the solution is based on AWS sample 08-okta-xaa-delegation-chain, but its snippets and configurations are conceptual rather than a complete deployable solution, and the code is for non-production use. Part 2 is described as extending the approach to AgentCore Gateway, Cedar policies, a second agent on Amazon EKS, and an audit trail in Amazon CloudWatch.