Remote Code Execution in MLflow Statsmodels _load_model (CVE‑2024‑37052) Threatens ML Pipelines
What It Is — A deserialization flaw in the _load_model function of the open‑source MLflow platform allows a remote attacker to execute arbitrary code in the context of the current user. Exploitation requires the victim to open a malicious file or visit a crafted web page that triggers the vulnerable function. The vulnerability is tracked as CVE‑2024‑37052, carries a CVSS 7.8 (AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H) rating, and was publicly disclosed on 29 Oct 2026. MLflow versions prior to 3.15.0 are affected; the issue was patched in MLflow 3.15.0.
Affected Products — MLflow (all versions < 3.15.0).
Why It Matters for Trust & Control Assurance — The flaw highlights the control objective of secure software development and configuration management—specifically, the need for rigorous input validation and evidence‑based monitoring of third‑party components. Verisq’s Control Mapping capability helps you map this gap to your framework of record and capture defensible audit evidence.
Recommended Actions
- Upgrade MLflow to 3.15.0 or later.
- Add runtime monitoring to detect unexpected process behavior after model loading.
- Enforce strict validation and sanitisation of all model artifacts before invoking
_load_model. - Maintain an inventory of open‑source dependencies and their patch status.