MLflow’s Experiment‑scoped Label Schema CRUD API lacks proper authorization checks in releases prior to commit 4666cffc7912ea606d592fc38d6a75e2935f65e7. An attacker who can reach the public API can create, modify, or delete label schema definitions that determine how experiments are tagged. This flaw encompasses multiple authorization bypass weaknesses, including missing authorization in the generic functions (CWE‑863), missing authorization for specific actions (CWE‑862), and a user‑controlled key bypass (CWE‑639).

The vulnerability affects any MLflow deployment running a commit prior to 4666cffc7912ea606d592fc38d6a75e2935f65e7. Subsequent releases are believed to incorporate authentication handlers for the label‑schema endpoints, as indicated by the GitHub issue discussion, but the exact commit that introduced the fix is not yet publicly documented.

The CVSS score of 2.3 indicates low overall severity, and the EPSS score of less than 1% points to a very low probability of exploitation. The issue is not catalog. The attack vector is remote, with high complexity and difficult exploitability inferred from the description. Although the flaw does not enable full system compromise, it permits an unauthenticated or improperly authorized user to alter experiment labeling schemas, potentially disrupting workflows and confidentiality of experiment metadata.