NewApp Monitoring is liveSign up free
AgentbeamBeta
AI agent vulnerability · MLflow

CVE-2024-27132: Cross-site Scripting in MLFlow

criticalCVSS 3.1 9.6GHSA-6749-m5cp-6cg7

MLflow before 2.10.0 does not sanitize template variables when it runs an MLflow Recipe. If you run an untrusted recipe, its content becomes cross-site scripting in the notebook output. In Jupyter Notebook this becomes code execution on the user's machine. Upgrade to mlflow 2.10.0.

Published Feb 24, 2024 · Updated Mar 15, 2026 · Analysis Nov 2, 2026

What an attacker can do

MLflow Recipes are configured with template files and variables. MLflow inserted those variables into rendered output without sanitization. A recipe built by an attacker can carry script in a template variable. When a data scientist runs the recipe and views its output, the script runs in their browser.

In Jupyter Notebook the output is rendered inside the notebook page, which can drive the notebook's kernel. JFrog and the advisory describe this as client-side remote code execution. JFrog's proof of concept loads the recipe with Recipe(profile="local"), runs it and inspects or retrieves artifacts. The page does not name the template variable or show how the script reaches the kernel.

The attacker must get you to run their recipe, for example from a shared repository, a tutorial or a pull request. That matches the user-interaction flag in the CVSS vector. The scope is marked changed, because the script escapes the recipe into the notebook session. It runs with your account's access to data, credentials and cloud tokens.

Who is affected

mlflow below 2.10.0. JFrog's range is 2.9.2 and earlier. The exposure arises when you run a recipe from an untrusted source, mainly inside Jupyter Notebook.

Check the installed version:

pip show mlflow

A sibling flaw with the same affected range, CVE-2024-27133, concerns untrusted datasets rather than template variables.

mlflow

Ecosystem
PyPI
Affected
< 2.10.0
Fixed in
2.10.0
Check
pip show mlflow

How to fix it

Upgrade to mlflow 2.10.0:

pip install --upgrade "mlflow>=2.10.0"

Restart the notebook kernel afterwards. The fix is pull request 10873 in the MLflow repository. JFrog lists no mitigation. Until you upgrade, do not run recipes from sources you do not trust.

How to tell if you were hit

Review recipe files you received from others for template variables or profile values that contain script tags, javascript: strings or HTML. In the notebook, look at the rendered output of Recipe runs and inspect calls for content you did not expect. After running an untrusted recipe on a vulnerable version, check shell history, new files, and outbound connections from the Jupyter server for the same time window. Rotate tokens that were present in the notebook session.

Original advisory

From the GitHub Advisory Database (CC BY 4.0).

Insufficient sanitization in MLflow leads to XSS when running an untrusted recipe.

This issue leads to a client-side RCE when running an untrusted recipe in Jupyter Notebook.

The vulnerability stems from lack of sanitization over template variables.

Questions

Is this exploitable without Jupyter?

The XSS needs a browser to render the output. The advisory says the full client-side code execution happens when the untrusted recipe is run in Jupyter Notebook.

How does CVE-2024-27132 differ from CVE-2024-27133?

This one is caused by unsanitized template variables in an untrusted recipe. CVE-2024-27133 is caused by unsanitized dataset table fields in an untrusted dataset. Both are fixed in mlflow 2.10.0.

Does the MLflow tracking server need to be exposed?

The advisory does not say so. The trigger is running an untrusted recipe, not reaching a server. The CVSS vector still reads network, because the malicious file can arrive remotely.

References

How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology

More MLflow vulnerabilities

All MLflow entries