AI Threat Alert
CVE-2025-15036: MLflow: path traversal enables sandbox escape, file overwrite
GHSA-vhcx-3pq2-4fvc CRITICAL CISA: ATTENDMLflow versions before v3.7.0 allow an attacker who controls a tar.gz artifact to overwrite arbitrary files and potentially escape the sandbox in multi-tenant or shared cluster environments (e.g., Databricks Connect). If your ML platform ingests user-supplied or third-party model artifacts via MLflow, treat this as HIGH severity and patch to v3.7.0 immediately. In shared environments where multiple teams or tenants load artifacts, this is a lateral movement and privilege escalation vector—prioritize accordingly.
What is the risk?
Although no CVSS score has been assigned yet, this vulnerability carries high effective risk in enterprise MLOps environments. The exploit primitive—crafting a malicious tar.gz—is trivial and well-documented. Impact is severe in multi-tenant ML platforms where sandbox isolation is a security boundary. Single-tenant, air-gapped deployments where artifact provenance is tightly controlled face lower—but not zero—risk. The absence of CISA KEV listing and EPSS score reflects recency, not low severity. Shared cluster environments using Databricks Connect are the highest-risk deployment pattern.
What systems are affected?
| Package | Ecosystem | Vulnerable Range | Patched |
|---|---|---|---|
| MLflow | pip | < 3.9.0rc0 | 3.9.0rc0 |
Do you use MLflow? You're affected.
How severe is it?
What is the attack surface?
What should I do?
5 steps-
PATCH
Upgrade MLflow to v3.7.0 or later immediately—this is the definitive fix. Reference commit: 3bf6d81ac4d38654c8ff012dbd0c3e9f17e7e346.
-
VERIFY PROVENANCE
Until patched, enforce strict provenance controls on all tar.gz artifacts ingested by MLflow—only accept artifacts from trusted, cryptographically signed sources.
-
RESTRICT ARTIFACT SOURCES
Disable ingestion of externally-sourced model artifacts in multi-tenant clusters if patching is delayed.
-
DETECT
Search logs for unexpected file writes outside expected MLflow working directories during artifact extraction. Monitor for new files created in /etc, /usr, or home directories during MLflow model load operations.
-
AUDIT DEPLOYMENTS
Identify all Databricks Connect and MLflow deployments using
dbconnect_artifact_cache—prioritize shared/multi-tenant clusters.
What does CISA's SSVC say?
Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.
How is it classified?
Which compliance frameworks are affected?
This CVE is relevant to:
Frequently Asked Questions
What is CVE-2025-15036?
MLflow versions before v3.7.0 allow an attacker who controls a tar.gz artifact to overwrite arbitrary files and potentially escape the sandbox in multi-tenant or shared cluster environments (e.g., Databricks Connect). If your ML platform ingests user-supplied or third-party model artifacts via MLflow, treat this as HIGH severity and patch to v3.7.0 immediately. In shared environments where multiple teams or tenants load artifacts, this is a lateral movement and privilege escalation vector—prioritize accordingly.
Is CVE-2025-15036 actively exploited?
No confirmed active exploitation of CVE-2025-15036 has been reported, but organizations should still patch proactively.
How to fix CVE-2025-15036?
1. PATCH: Upgrade MLflow to v3.7.0 or later immediately—this is the definitive fix. Reference commit: 3bf6d81ac4d38654c8ff012dbd0c3e9f17e7e346. 2. VERIFY PROVENANCE: Until patched, enforce strict provenance controls on all tar.gz artifacts ingested by MLflow—only accept artifacts from trusted, cryptographically signed sources. 3. RESTRICT ARTIFACT SOURCES: Disable ingestion of externally-sourced model artifacts in multi-tenant clusters if patching is delayed. 4. DETECT: Search logs for unexpected file writes outside expected MLflow working directories during artifact extraction. Monitor for new files created in /etc, /usr, or home directories during MLflow model load operations. 5. AUDIT DEPLOYMENTS: Identify all Databricks Connect and MLflow deployments using `dbconnect_artifact_cache`—prioritize shared/multi-tenant clusters.
What systems are affected by CVE-2025-15036?
This vulnerability affects the following AI/ML architecture patterns: Training pipelines, MLOps platforms, Model serving, Multi-tenant ML environments, CI/CD ML pipelines.
What is the CVSS score for CVE-2025-15036?
CVE-2025-15036 has a CVSS v3.1 base score of 9.6 (CRITICAL). The EPSS exploitation probability is 0.54%.
What is the AI security impact?
Affected AI Architectures
MITRE ATLAS Techniques
AML.T0010.001 AI Software AML.T0011.000 Unsafe AI Artifacts AML.T0035 AI Artifact Collection AML.T0049 Exploit Public-Facing Application AML.T0105 Escape to Host Compliance Controls Affected
What are the technical details?
Original Advisory
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.
Exploitation Scenario
An attacker operating in a multi-tenant ML environment (e.g., a shared Databricks workspace) crafts a malicious tar.gz file containing path traversal sequences in member filenames (e.g., `../../etc/cron.d/backdoor`). They publish this as a model artifact to a shared MLflow Model Registry—either by compromising a low-privilege account, exploiting a misconfigured registry, or via a poisoned upstream dependency. When a victim user or automated pipeline loads the model via `mlflow.pyfunc.load_model()`, MLflow's artifact cache mechanism extracts the tar.gz without validating member paths. The malicious file is written outside the sandbox directory, potentially overwriting a cron job, SSH authorized_keys, or Python site-packages to achieve persistent code execution with elevated privileges on the shared cluster.
Weaknesses (CWE)
CWE-29 — Path Traversal: '..filename': The product uses external input to construct a pathname that should be within a restricted directory, but it does not properly neutralize '\..\filename' (leading backslash dot dot) sequences that can resolve to a location that is outside of that directory.
- [Implementation] Assume all input is malicious. Use an "accept known good" input validation strategy, i.e., use a list of acceptable inputs that strictly conform to specifications. Reject any input that does not strictly conform to specifications, or transform it into something that does. When performing input validation, consider all potentially relevant properties, including length, type of input, the full range of acceptable values, missing or extra inputs, syntax, consistency across related fields, and conformance to business rules. As an example of business rule logic, "boat" may be syntactically valid because it only contains alphanumeric characters, but it is not valid if the input is only expected to contain colors such as "red" or "blue." Do not rely exclusively on looking for malicious or malformed inputs. This is likely to miss at least one undesirable input, especially if the code's environment changes. This can give attackers enough room to bypass the intended validation. However, denylis
- [Implementation] Inputs should be decoded and canonicalized to the application's current internal representation before being validated (CWE-180). Make sure that the application does not decode the same input twice (CWE-174). Such errors could be used to bypass allowlist validation schemes by introducing dangerous inputs after they have been checked.
Source: MITRE CWE corpus.
CVSS Vector
CVSS:3.0/AV:N/AC:L/PR:N/UI:R/S:C/C:H/I:H/A:H References
Timeline
Related Vulnerabilities
CVE-2025-15379 10.0 MLflow: RCE via unsanitized model dependency specs
Same package: mlflow CVE-2023-3765 10.0 MLflow: path traversal allows arbitrary file read
Same package: mlflow CVE-2026-2635 9.8 mlflow: security flaw enables exploitation
Same package: mlflow CVE-2023-1177 9.8 MLflow: path traversal allows arbitrary file read/write
Same package: mlflow CVE-2023-2780 9.8 MLflow: path traversal allows arbitrary file read/write
Same package: mlflow