Security · MLflow MLflow
MLflow Command Injection Flaw (CVE-2025-15379) Fixed in Version 3.8.2
MLflow's model serving container initialization contains a command injection vulnerability (CVE-2025-15379) in the `_install_model_dependencies_to_env()` function, where dependency entries from a model artifact's `python_env.yaml` are interpolated into a shell command without sanitization when deploying with `env_manager=LOCAL`, allowing a malicious model artifact to execute arbitrary commands on the deploying system. The issue affects version 3.8.0 and is fixed in 3.8.2; separately, an update to CVE-2025-14813 covers a broken/risky cryptographic algorithm issue in Bouncy Castle BC-JAVA (bcprov) G3413CTRBlockCipher affecting 1.59 before 1.80.2, 1.81 before 1.81.1, and 1.82 before 1.84.
Read the original coverage at National Vulnerability Database (NVD)
Captured 8 weeks ago