CVE-2025-15031: MLflow Path Traversal enables

CISO Take

MLflow is the de facto MLOps backbone in enterprise AI pipelines — this path traversal vulnerability turns any trusted user with artifact upload rights into a potential RCE threat actor. A crafted model bundle can overwrite arbitrary files on the MLflow server, including cron jobs, SSH keys, or application configs. Immediately restrict artifact uploads to authenticated internal users only and sandbox MLflow extraction processes until a patch is available.

What is the risk?

HIGH in any multi-tenant or externally-accessible MLflow deployment. The attack complexity is low: any actor with artifact upload capability can exploit it without elevated privileges. No CVSS score assigned yet, but the combination of CWE-22 (path traversal) + arbitrary file write + potential RCE in a widely-deployed AI infrastructure component warrants treating this as Critical. Risk degrades to MEDIUM only in air-gapped, single-tenant deployments with strict access control and no external artifact ingestion. Container-based deployments with host mounts face container escape risk, compounding severity.

What systems are affected?

Package	Ecosystem	Vulnerable Range	Patched
MLflow	pip	< 3.9.0rc0	`3.9.0rc0`
26.6K OpenSSF 5.6 655 dependents Pushed 4d ago 31% patched ~51d to patch Full package profile →

Do you use MLflow? You're affected.

How severe is it?

CVSS 3.1

9.1 / 10

EPSS

0.7%

chance of exploitation in 30 days

Higher than 49% of all CVEs

Source: EPSS v3 — FIRST.org

Exploitation Status

Exploit Available

Exploitation: MEDIUM

Sophistication

Advanced

Exploitation Confidence

medium

○ CISA SSVC: Public PoC

Composite signal derived from CISA KEV, VulnCheck KEV, CISA SSVC, EPSS, Metasploit, Exploit-DB, trickest/cve, Nuclei templates, and inthewild.io exploitation reports.

What is the attack surface?

AV Network

AC Low

PR None

UI None

S Unchanged

C High

I High

A None

What should I do?

6 steps

IMMEDIATE

Restrict MLflow artifact upload access to authenticated, known-good internal accounts only. Revoke or audit any external/partner upload credentials.
WORKAROUND

Run MLflow artifact extraction in ephemeral, isolated containers with no host mounts and a read-only filesystem outside the extraction target directory.
DETECTION

Scan existing artifact storage for tar.gz/tar archives containing path traversal sequences (../ or absolute paths starting with /). Use: python3 -c "import tarfile, sys; t=tarfile.open(sys.argv[1]); [print(m.name) for m in t.getmembers() if m.name.startswith('/') or '../' in m.name]" <artifact>.
MONITORING

Alert on file creation events in system directories (e.g., /etc, /usr, ~/.ssh) from MLflow process lineage.
PATCH

Track the huntr.com disclosure for official fix; prioritize patching as soon as MLflow releases a remediated version.
VERIFY

After patching, confirm the fix replaces tarfile.extractall with path-sanitized extraction using tarfile.extractall with a custom filter or manual member validation.

What does CISA's SSVC say?

Decision Attend

Exploitation poc

Automatable No

Technical Impact total

Source: CISA Vulnrichment (SSVC v2.0). Decision based on the CISA Coordinator decision tree.

How is it classified?

Supply Chain Model Poisoning Code Execution Framework Model Training Data AML.T0010.001 - AI Software AML.T0010.003 - Model AML.T0011.000 - Unsafe AI Artifacts AML.T0018.002 - Embed Malware AML.T0049 - Exploit Public-Facing Application AML.T0105 - Escape to Host

Which compliance frameworks are affected?

This CVE is relevant to:

EU AI Act

Article 15 - Accuracy, Robustness and Cybersecurity Article 9 - Risk Management System

ISO 42001

A.6.1.2 - AI Supply Chain Management A.8.4 - AI System Security

NIST AI RMF

GOVERN 1.7 - Policies and procedures are in place for third-party AI risks MANAGE 2.2 - Mechanisms are in place to respond to and recover from AI risks

OWASP LLM Top 10

LLM05:2025 - Insecure Output Handling / Supply Chain Vulnerabilities

Frequently Asked Questions

What is CVE-2025-15031?

MLflow is the de facto MLOps backbone in enterprise AI pipelines — this path traversal vulnerability turns any trusted user with artifact upload rights into a potential RCE threat actor. A crafted model bundle can overwrite arbitrary files on the MLflow server, including cron jobs, SSH keys, or application configs. Immediately restrict artifact uploads to authenticated internal users only and sandbox MLflow extraction processes until a patch is available.

Is CVE-2025-15031 actively exploited?

No confirmed active exploitation of CVE-2025-15031 has been reported, but organizations should still patch proactively.

How to fix CVE-2025-15031?

1. IMMEDIATE: Restrict MLflow artifact upload access to authenticated, known-good internal accounts only. Revoke or audit any external/partner upload credentials. 2. WORKAROUND: Run MLflow artifact extraction in ephemeral, isolated containers with no host mounts and a read-only filesystem outside the extraction target directory. 3. DETECTION: Scan existing artifact storage for tar.gz/tar archives containing path traversal sequences (`../` or absolute paths starting with `/`). Use: `python3 -c "import tarfile, sys; t=tarfile.open(sys.argv[1]); [print(m.name) for m in t.getmembers() if m.name.startswith('/') or '../' in m.name]" <artifact>`. 4. MONITORING: Alert on file creation events in system directories (e.g., /etc, /usr, ~/.ssh) from MLflow process lineage. 5. PATCH: Track the huntr.com disclosure for official fix; prioritize patching as soon as MLflow releases a remediated version. 6. VERIFY: After patching, confirm the fix replaces `tarfile.extractall` with path-sanitized extraction using `tarfile.extractall` with a custom filter or manual member validation.

What systems are affected by CVE-2025-15031?

This vulnerability affects the following AI/ML architecture patterns: MLOps platforms, training pipelines, model registries, model serving, CI/CD ML pipelines, multi-tenant ML platforms, automated model evaluation workflows.

What is the CVSS score for CVE-2025-15031?

CVE-2025-15031 has a CVSS v3.1 base score of 9.1 (CRITICAL). The EPSS exploitation probability is 0.71%.

What is the AI security impact?

Affected AI Architectures

MLOps platformstraining pipelinesmodel registriesmodel servingCI/CD ML pipelinesmulti-tenant ML platformsautomated model evaluation workflows

MITRE ATLAS Techniques

AML.T0010.001 AI Software

AML.T0010.003 Model

AML.T0011.000 Unsafe AI Artifacts

AML.T0018.002 Embed Malware

AML.T0049 Exploit Public-Facing Application

AML.T0105 Escape to Host

Compliance Controls Affected

EU AI Act: Article 15, Article 9

ISO 42001: A.6.1.2, A.8.4

NIST AI RMF: GOVERN 1.7, MANAGE 2.2

OWASP LLM Top 10: LLM05:2025

What are the technical details?

Original Advisory

A vulnerability in MLflow's pyfunc extraction process allows for arbitrary file writes due to improper handling of tar archive entries. Specifically, the use of `tarfile.extractall` without path validation enables crafted tar.gz files containing `..` or absolute paths to escape the intended extraction directory. This issue affects the latest version of MLflow and poses a high/critical risk in scenarios involving multi-tenant environments or ingestion of untrusted artifacts, as it can lead to arbitrary file overwrites and potential remote code execution.

Exploitation Scenario

An adversary with access to an MLflow tracking server — via compromised internal credentials, a malicious insider, or an upstream supply chain compromise — uploads a pyfunc model bundle packaged as a crafted tar.gz file. The archive contains entries with path traversal sequences such as `../../etc/cron.d/mlflow-backdoor` or `../../root/.ssh/authorized_keys`, each containing attacker-controlled content. When MLflow processes this artifact during model registration, loading, or automated evaluation in a CI/CD pipeline, `tarfile.extractall` writes these files to their traversed destinations without validation. On a bare-metal MLflow server, this achieves persistent RCE via cron or SSH key injection. On a Kubernetes pod with host path mounts (common in ML training setups), the traversal escapes the container and achieves host-level code execution — full cluster compromise is then a lateral movement step away.

Weaknesses (CWE)

CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal') Primary CWE-22 Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

CWE-22 — Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal'): The product uses external input to construct a pathname that is intended to identify a file or directory that is located underneath a restricted parent directory, but the product does not properly neutralize special elements within the pathname that can cause the pathname to resolve to a location that is outside of the restricted 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
[Architecture and Design] For any security checks that are performed on the client side, ensure that these checks are duplicated on the server side, in order to avoid CWE-602. Attackers can bypass the client-side checks by modifying values after the checks have been performed, or by changing the client to remove the client-side checks entirely. Then, these modified values would be submitted to the server.

Source: MITRE CWE corpus.