CVE-2026-33865

Affected versions of the mlflow package are vulnerable to Stored Cross-Site Scripting due to unsafe parsing of YAML-based MLmodel artifacts when rendered in the web interface. The web UI processes attacker-supplied fields from uploaded ML model files without adequate output sanitization, allowing embedded payloads to be emitted into the rendered artifact view and executed in the browser of any user who opens the artifact. An authenticated attacker who uploads a malicious ML model file can trigger script execution in the context of a victim's session, enabling session hijacking or actions performed on the victim's behalf.