Loading vulnerability…
CVE-2025-11201 is an unauthenticated directory traversal vulnerability in MLflow Tracking Server affecting the model creation process. The flaw is caused by improper validation of a user-supplied file path, specifically the source parameter used by model creation endpoints, before it is used in file operations. By supplying traversal sequences, a remote attacker can escape the intended artifact directory and write files to attacker-chosen locations on the server filesystem. This arbitrary file write can be turned into remote code execution if the attacker places a Python-loaded file, such as a module or path configuration file, in a location that will be processed by the Python runtime. Code execution occurs in the security context of the MLflow service account.
Mallory correlates every CVE against your assets, your vendors, and active adversary campaigns. Know which vulnerabilities matter for you, not just which ones are loud.
What it means. What to do now. Patch path, mitigations, and the assume-compromise checklist.
What an attacker gets, and what they’ve been doing with it.
If you can’t patch tonight, do this now.
Patch, then assume compromise.
1 valid exploit after Mallory filtered fakes, detection scripts, and README-only repos.
This two-file repository contains a README and a standalone Python proof of concept for CVE-2025-11201 affecting MLflow versions before 3.0.0 (tested against 2.22.0 on Ubuntu 22.04). poc.py uses the MLflow REST API to create an experiment and run, upload an MLmodel manifest and evil_loader.py through the artifact proxy, create a registered model, and register a version pointing to the uploaded artifacts. It adds the public mlflow.prompt.is_prompt=true tag, which the PoC alleges causes source validation to be skipped. When a downstream client loads models:/<name>/<version>, MLflow downloads the artifact and imports evil_loader, resulting in Python code execution as the user running that client. The bundled payload only writes a /tmp marker with UID/GID details, but the loader-module mechanism permits replacement with arbitrary Python code. No exploit framework is used.
Products and vendors Mallory has correlated with this vulnerability. Open in Mallory to drill down to specific CPE configurations and version ranges.
Vendor-confirmed product mapping. Mallory continuously reconciles this list against your asset inventory.
9 sources tracked across advisories, community write-ups, and news. New activity surfaces here as Mallory finds it.
A critical MLflow unauthenticated path traversal leading to RCE in model creation, referenced as historical context and comparison to similar hardening patterns.
A directory traversal vulnerability in MLflow Tracking Server's model creation process that can allow arbitrary file write and lead to remote code execution without authentication in exposed environments.
Query your assets running an affected version, and investigate the blast radius.
Every observed campaign linking this CVE to a named adversary.
Malware families riding this exploit, with evidence and IOCs.
YARA, Sigma, Snort, and vendor rules, auto-deployed to your SIEM.
Cross-references every affected SKU, including bundled OEM variants.
Community discussion across Reddit, Mastodon, and other social sources.