CVE-2025-0453: MLflow Uncontrolled Resource Consumption vulnerability
MLflow 2.17.2 and earlier lets an unauthenticated attacker knock the tracking server over through the /graphql endpoint. Large batches of queries that each request all runs of an experiment tie up every worker, so the server stops answering other requests. The advisory lists no patched version.
Published Mar 20, 2025 · Updated Oct 15, 2025 · Analysis Nov 1, 2026
What an attacker can do
The /graphql endpoint accepts batched queries without a limit on their size or cost. A query that asks for all runs in an experiment is expensive. Repeating it many times in one batch keeps the allocated MLflow workers busy. Legitimate requests, such as logging metrics or loading the UI, then wait or fail.
The attacker needs network access to the server and no account, as the vector shows. Attack complexity is rated high, perhaps because the effect depends on how many workers are set and how large the experiment is. The impact is availability only. Data is not read or changed.
For a team, the risk is stalled training pipelines and tracking that depends on the server. Its EPSS score is 11.3 percent at the 96th percentile, higher than most advisories of this class. No source says it is exploited in the wild.
Who is affected
mlflow (pip) 2.17.2 and earlier. The advisory names no fixed release. Check your version:
pip show mlflowExposed servers are those whose /graphql route can be reached by untrusted clients. A server on a private network with limited access carries less risk.
mlflow
- Ecosystem
- PyPI
- Affected
- <= 2.17.2
- Fixed in
- No patched version yet
- Check
pip show mlflow
How to fix it
No patched version is listed, and the advisory gives no mitigation. Check the MLflow release notes for newer releases than 2.17.2 and upgrade.
Until then, keep the tracking server off the open internet, require authentication in front of it, and use a reverse proxy to rate-limit or restrict requests to /graphql. The sources do not describe these steps, so test them before you rely on them.
How to tell if you were hit
Look in the server access logs for bursts of POST requests to /graphql, especially large bodies that contain many repeated queries for the same experiment's runs. Watch for worker saturation: high CPU, requests queuing, and timeouts on other routes such as the UI or the REST API. Note the source addresses of the bursts.
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
In mlflow/mlflow version 2.17.2, the /graphql endpoint is vulnerable to a denial of service attack. An attacker can create large batches of queries that repeatedly request all runs from a given experiment. This can tie up all the workers allocated by MLFlow, rendering the application unable to respond to other requests. This vulnerability is due to uncontrolled resource consumption.
Questions
- Does CVE-2025-0453 need authentication?
The CVSS vector requires no privileges and no user interaction. Anyone who can reach /graphql on a default setup can send the queries.
- Does it leak or change data?
No. Only availability is affected. The vector rates confidentiality and integrity impact as none.
- Is there a patched MLflow version?
The advisory lists none. It marks 2.17.2 and earlier as affected, so check the release notes for newer versions.
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 entriesCVE-2026-116690
MLflow vulnerable to arbitrary file read on unauthenticated servers via prompt model version source
high7.5Oct 27, 2026CVE-2026-116691
MLflow CrewAI autologging records LLM provider API keys in logged trace spans
medium6.5Oct 27, 2026CVE-2026-116692
MLflow vulnerable to remote code execution on unauthenticated servers via third-party scorer deserialization importing caller-placed artifact modules
critical9.8Oct 27, 2026CVE-2026-115886
MLflow: SSRF on unauthenticated MLflow servers via third-party scorer kwargs in the scorer invoke, register, and GenAI evaluate endpoints
medium6.5Oct 23, 2026CVE-2026-53995
MLflow FileStore trace IDs allow path traversal and arbitrary file writes
high8.2Oct 20, 2026GHSA-gqvg-gmmx-x4hm
MLFLOW_ALLOW_PICKLE_DESERIALIZATION=False safety control bypassed by mlflow.statsmodels flavor — RCE via crafted model artifact
high8.8Sep 1, 2026
