Junglewise Threat Intelligence

CVE-2025-15379: MLflow command injection in model serving container initialization

CVE-2025-15379 · Severity: critical · CVSS 9.8 · Published 2026-03-30

Technologies: LF Projects Mlflow, Red Hat OpenShift AI (RHOAI). Vendors: Red Hat, PyPI.

Executive brief

MLflow is an open-source platform used to manage the machine learning lifecycle, including model deployment. A security flaw allows an attacker to execute malicious commands on the server by providing a specially crafted model file. This could lead to a complete takeover of the system hosting the machine learning models, potentially exposing sensitive data or disrupting operations.

Technical details

A command injection vulnerability exists in MLflow's model serving container initialization code within the `_install_model_dependencies_to_env()` function. When a model is deployed using the `env_manager=LOCAL` configuration, the application reads dependency specifications from the model artifact's `python_env.yaml` file. These specifications are directly interpolated into a shell command without proper sanitization or escaping. An attacker can exploit this by supplying a malicious model artifact containing shell metacharacters in the dependency fields, leading to arbitrary command execution with the privileges of the MLflow process. The issue is fixed in version 3.8.2 by implementing proper shell escaping using `shlex`.

Affected products

  • LF Projects MLflow 3.8.0 to 3.8.1
  • Red Hat OpenShift AI (RHOAI)

Timeline

  • 2025-12-24: patched: Fix committed to GitHub repository
  • 2026-03-30: disclosed: Initial disclosure by huntr.dev and NVD publication

References

Related threats