A medium-severity vulnerability in MLflow's built-in basic-auth plugin allows any authenticated user to inject arbitrary dataset records into another user's run via the LogInputs endpoint, according to a GitHub security advisory1. Tracked as CVE-2026-69146, the flaw exists because the LogInputs proto handler is absent from the BEFORE_REQUEST_HANDLERS map, causing the before-request authorization hook to skip validation entirely. Standard write endpoints such as log-metric correctly return HTTP 403. The vulnerability affects MLflow versions below 3.15.0 and is patched in version 3.15.0.
MLflow Patches Auth Bypass in LogInputs Endpoint (CVE-2026-69146)
CVE-2026-69146 discloses an authorization bypass in MLflow's basic-auth plugin allowing authenticated users to write dataset records to any run. Patched…