MLFLOW · CVE-2025-15036 · AI agents & MCP
FIX → 3.9.0rc0 · 3.9.0
MLflow's artifact cache extracts tar.gz archives without checking member paths, so a crafted archive can write files outside the target directory. An attacker who controls the archive can overwrite arbitrary files or gain elevated privileges, escaping the sandbox in shared clusters.
first public Mar 30, 2026 · 1 public exploit · 2 sources
Check it with your coding agent
It reads this page and tells you if your code or app is affected.
Teams running MLflow below 3.9.0rc0 that load pyfunc models or artifacts from archives they do not fully trust, especially in multi-tenant or shared cluster environments. Single-user local setups that only load trusted artifacts are less exposed. unknown whether the vulnerable path is reached in every deployment.
Not affected if single-user, trusted-artifacts-only
Who cares ml-platform, mlflow, python, shared-clusters
pip show mlflow; versions below 3.9.0rc0 are affected.extract_archive_to_dir in mlflow/pyfunc/dbconnect_artifact_cache.py in your deployed code.A path traversal vulnerability exists in the extract_archive_to_dir function within the mlflow/pyfunc/dbconnect_artifact_cache.py file of the mlflow/mlflow repository. This vulnerability, present in versions before v3.7.0, arises due to the lack of validation of tar member paths during extraction. An attacker with control over the tar.gz file can exploit this issue to overwrite arbitrary files or gain elevated privileges, potentially escaping the sandbox directory in multi-tenant or shared cluster environments.
Are you affected?
Ask from here. InfraStack checks the affected versions against your stack and answers in your inbox.
InfraStack checks the affected versions against your stack and answers in Slack, with the evidence and the fix.
InfraStack checks it against your stack and answers on your report, with the evidence and the fix.
Upgrade to 3.9.0rc0 · 3.9.0
mlflow moves to 3.9.0rc0 or later.GitHub Advisory Database, CC-BY 4.0 · NVD data is provided by NIST and is public domain. Posts quoted belong to their authors.