CVE-2024-27132

The core impact of CVE-2024-27132 lies in the ability of an attacker to inject malicious JavaScript code into a recipe executed within MLflow. Because recipes are often run within Jupyter Notebooks, this injected code can execute with the privileges of the user running the notebook. Successful exploitation could allow an attacker to steal sensitive data, such as API keys, database credentials, or model artifacts. More severely, it could enable the attacker to execute arbitrary commands on the underlying system, leading to complete system compromise. The RCE aspect significantly elevates the risk, moving beyond simple information disclosure to full system control. This vulnerability shares similarities with other XSS vulnerabilities that have been exploited to achieve code execution in similar environments.

Organizations heavily reliant on MLflow for machine learning model management and deployment are at significant risk. Specifically, teams using MLflow within Jupyter Notebook environments, particularly those dealing with sensitive data or running recipes from external sources, should prioritize remediation. Shared hosting environments where multiple users execute MLflow recipes are also at increased risk.

• python / mlflow:

import mlflow
# Check MLflow version
print(mlflow.__version__)
# Monitor Jupyter Notebook logs for unusual JavaScript execution

• generic web:

curl -I <mlflow_endpoint> | grep -i 'x-content-type-options'

• generic web:

curl -I <mlflow_endpoint> | grep -i 'content-security-policy'

1. 2024-02-23Disclosure

   disclosure

1. आरक्षित2024-02-20
2. प्रकाशित2024-02-23
3. संशोधित2025-04-08
4. EPSS अद्यतन2026-04-02

The primary mitigation for CVE-2024-27132 is to upgrade MLflow to version 2.10.0 or later, which includes the necessary sanitization fixes. If immediate upgrading is not possible, restrict the execution of untrusted recipes to isolated environments with limited privileges. Consider implementing a Web Application Firewall (WAF) with rules to detect and block malicious JavaScript payloads targeting MLflow endpoints. Carefully review and validate all recipes before execution, particularly those sourced from external or untrusted origins. There are no specific Sigma or YARA rules readily available, but monitoring for unusual JavaScript execution within Jupyter Notebooks is recommended.