ZDI-26-105 MLflow Tracking Server Vulnerability

TL;DR

A remote attacker can exploit an unauthenticated RCE vulnerability in MLflow's Artifact Handler, allowing them to execute code with elevated privileges.

What happened

["MLflow Tracking Server's artifact handler allows for remote code execution without authentication", 'Attackers can traverse directories and execute arbitrary code']

Why it matters for ops

['Failure to sanitize input leads to directory traversal vulnerability', 'Lack of proper access controls allows RCE']

Mitigation

• Apply security patches for MLflow version >= 2.7.1
• Implement strict input validation and access control policies

Action items

• Upgrade MLflow Tracking Server to mitigate vulnerability
• Monitor logs and network traffic for suspicious activity

Detection IOCs

• Unusual network activity related to the artifact handler
• Unexpected changes or additions to system files

Source link

http://www.zerodayinitiative.com/advisories/ZDI-26-105/