Incorrect authorization in auth0-ai-js - CVE-2026-107372
Published: November 11, 2026
Vulnerability details
The vulnerability allows a remote user to disclose document content that authorization checks deny them access to through generated responses.
The vulnerability exists due to incorrect authorization in the auth0-ai-js FGA retriever when filtering documents during retrieval-augmented generation with Auth0 FGA conditional checks. A remote user can trigger retrieval in which authorization checks under different conditions are not distinguished, causing denied documents to be included in the model context and disclosing their content through generated responses.
Exploitation requires a single retrieval to include multiple documents that share the same object, user, and relation but differ by authorization context.