MLflow vulnerable to remote code execution on unauthenticated servers via third-party scorer deserialization importing caller-placed artifact modules
What to do
- Upgrade to a fixed version: mlflow 3.17.0.
- Redeploy and restart long-running processes so they load the fixed code.
Affected packages
mlflow
- >= 3.12.0rc0, < 3.17.0
- 3.17.0
Check your project: pip show mlflow
What changed
Published in the GitHub Advisory Database as critical.