Loom authentication flaw exposes AI agent administration to remote attackers

A critical authentication bypass in Loom can give unauthenticated remote attackers administrative access to its AI agent control plane, Yahoo reported. Loom manages agents on Amazon Bedrock AgentCore Runtime and AWS Strands Agents. Versions before 1.6.1 return a fixed super-administrator identity when no identity provider is configured. Attackers can use that access to register malicious tool servers, steal integration credentials or change AWS access policies, potentially escalating privileges into the underlying cloud environment. The authentication fix requires version 1.6.1 or later. Two companion flaws involving token disclosure and server-side requests affect versions before 1.7.0; exposed installations also require credential rotation and a review of cloud activity logs.