Uncontrolled Resource Consumption in mlflow/mlflow

Vulnerability

Overview

CVE-2024-6838 describes a denial-of-service vulnerability in MLflow version 2.13.2. The software fails to enforce a limit on the length of experiment names or the artifact_location parameter when creating or renaming experiments. An attacker can supply a name consisting of a large number of integers, causing the MLflow UI panel to become unresponsive [1].

Exploitation

An attacker with access to the MLflow tracking server can exploit this by sending a crafted API request to create or rename an experiment with an excessively long name or a large artifact_location value. No authentication is required if the server is exposed, making the attack surface broad in deployments without proper access controls [3].

Impact

Successful exploitation leads to a denial-of-service condition where the MLflow UI becomes unresponsive, preventing legitimate users from viewing or managing experiments. This can disrupt machine learning workflows and degrade platform availability [1].

Mitigation

As of the publication date, no official patch has been announced for this vulnerability. Users are advised to restrict network access to the MLflow tracking server and monitor for unusual experiment creation activity. The issue was reported via the Huntr bug bounty program [3].