CVE-2023-2780

Deeper analysis

The vulnerability is a path traversal flaw (CWE-29) that accepts traversal sequences such as '\..\filename' and affects the open-source MLflow machine-learning platform in all versions prior to 2.3.1. The issue received a CVSS 3.1 base score of 9.8, reflecting network-accessible attack vectors that require no authentication or user interaction.

An unauthenticated remote attacker can supply crafted paths to read or write arbitrary files on the server, resulting in full compromise of confidentiality, integrity, and availability. Because the flaw is exposed through MLflow's network interfaces, any reachable instance is potentially reachable by an external adversary.

The maintainers resolved the issue in commit fae77a525dd908c56d6204a4cef1c1c75b4e9857, which shipped with MLflow 2.3.1; the referenced huntr.dev report confirms the same remediation. Operators should upgrade immediately and verify that no untrusted path inputs reach the affected endpoints.

The associated EPSS score has reached a peak of 0.8777 (current value 0.8614), indicating sustained and substantial exploitation interest since disclosure. As an ML-lifecycle tool, the affected component is commonly deployed in data-science and production environments, increasing the practical impact of successful attacks.