Skip to content
MLflow Remote Code Execution Vulnerability Discovered

MLflow Remote Code Execution Vulnerability Discovered

First seen 28 Oct 2026, 04:45 UTC • • •Early signal · one source

Article Content

Browse articles
ThreatCluster AI
ThreatCluster •October 29, 2026 at 04:38 UTC
  • •MLflow versions 3.12.0 and 3.14.0 are vulnerable to remote code execution.
  • •Attackers can exploit the flaw via third-party scorer deserialization on unauthenticated servers.
  • •A proof of concept exists, demonstrating the vulnerability without causing harm.

A vulnerability in MLflow versions 3.12.0 and 3.14.0 allows remote code execution on unauthenticated servers via third-party scorer deserialization. The flaw enables attackers to upload malicious Python modules that can be executed with the privileges of the MLflow server process. An attacker can exploit this vulnerability by creating an experiment with a writable artifact location, uploading a malicious module, and triggering its execution through the MLflow API. This results in arbitrary code execution, potentially compromising the server. The vulnerability has been confirmed and a proof of concept has been released, although it is harmless. Users are advised to apply patches to mitigate this risk. The vulnerability is tracked under CVE-2026-XXXX, with no known reported at this time.

Start a free Starter trial for enhanced analysis

Ask AI about this cluster

Updated 17d ago How this analysis works

Timeline

2026-10-27
Vulnerability disclosed
MLflow's remote code execution vulnerability was publicly disclosed, detailing the exploit method and affected versions.
Github

More articles in this cluster (1)

Common questions

Which versions of MLflow are affected?
MLflow versions 3.12.0 and 3.14.0 are confirmed to be vulnerable.
Is there a patch available?
Yes, users are advised to update to the latest version of MLflow to mitigate the vulnerability.
What is the potential impact of this vulnerability?
The vulnerability allows for remote code execution, which can lead to full server compromise.