A critical security flaw, identified as CVE-2023-6018, has been uncovered in MLflow, An open source platform for the machine learning lifecycle. This vulnerability, disclosed by @kevin-mizu at Huntr, a bug bounty program for AI and ML, poses a severe risk as it allows unauthorized attackers to overwrite any file on the server where MLflow is hosted, without requiring authentication.
By exploiting this flaw, attackers gain the ability to manipulate files on the server system, enabling them to execute commands and potentially access sensitive data and model information. For instance, an attacker could compromise the system by overwriting files like
/home/<user>/.bashrc, leading to potential code execution and unauthorized access to critical data.
The ESS score has recently been escalated to 87.11% from 0.10%.
The affected software includes versions of pip/mlflow up to 2.8.1, as well as Lfprojects Mlflow.
For a comprehensive breakdown of the vulnerability, refer to https://huntr.com/bounties/7cf918b5-43f4-48c0-a371-4d963ce69b30/