Loading vulnerability…
MLflow contained a vulnerability where an attacker could upload arbitrary files to the server via a PUT request. This flaw allows unauthenticated or unauthorized users to place files of their choosing onto the server's filesystem, potentially leading to further exploitation such as remote code execution or data compromise.
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 three-file repository contains an MIT license, a detailed README, and one Python entry-point script, mlflow_pickle_rce.py. It is an operational, API-driven proof of concept for MLflow model-registry pickle deserialization leading to remote code execution, described as CVE-2024-37054-style and referencing CVE-2023-6015, CVE-2023-6018, and CVE-2024-37054. The script uses requests with HTTP Basic authentication to test MLflow API access, create or reuse a registered model, create an experiment run, upload an MLmodel manifest and malicious model.pkl, register a model version referencing those artifacts, and optionally transition the version to Production. The pickle executes os.system during unpickling and starts a Python reverse shell to configurable callback parameters. It does not identify a target model automatically or trigger model loading; a separate MLflow UI/API action, prediction service, or scheduled client job must load the registered model for execution to occur. Defaults include admin:password credentials, experiment ID 0, Production stage, and callback port 4444.
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.
No public activity tracked yet. Mallory keeps watching.
No public activity observed for this vulnerability.
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.