Executive brief
MLflow, a platform for managing machine learning lifecycles, contains a vulnerability in its labeling schema API. This flaw allows users to bypass intended access controls, potentially allowing unauthorized individuals to view or modify experiment data. While the attack is complex to execute, it could lead to unauthorized access to sensitive machine learning project configurations.
Technical details
A missing authorization vulnerability (CWE-862/CWE-863) exists in MLflow's Experiment-scoped Label Schema CRUD API. The vulnerability stems from a lack of authentication handlers in a specific pull request (PR #23607) that was merged or partially implemented without proper Role-Based Access Control (RBAC) enforcement. A remote attacker with low privileges can exploit this to bypass experiment-level access restrictions. The attack is considered high complexity because it requires specific knowledge of the unreleased or experimental API endpoints. Developers have indicated that authorization handlers were planned for a subsequent release.
Affected products
- MLflow MLflow up to 4666cffc7912ea606d592fc38d6a75e2935f65e7
Timeline
- 2026-06-28: disclosed: Public disclosure of the vulnerability and associated GitHub issue.
- 2026-06-28: advisory