CVE-2026-116690

MLflow: Arbitrary file read on unauthenticated MLflow servers via prompt model version source

Published Oct 27, 2026 CVSS 3.1 · 7.5 High EPSS 1.8% KEV · Not listed
Description
MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.10.0 until 3.17.0, the prompt branch in _create_model_version() accepts schemeless relative sources because path validation runs only when a source has a URL scheme, and mlflow.prompt.is_prompt is interpreted by tag presence rather than a true value. A remote caller can store a server-local path as the model version artifact location and use /api/2.0/mlflow/model-versions/get-artifact to read files accessible to the MLflow server process. Default unauthenticated servers require no privileges, while basic-auth deployments are exposed to authenticated users who can create registered models. This issue is fixed in version 3.17.0.

Analysis

AI-generated

Technical impact

The vulnerability allows an attacker to read any file on the MLflow server that the process can access. From versions 3.10.0 through 3.17.0, the prompt branch in the _create_model_version() function accepts schemeless relative paths as artifact locations. Because path validation is performed only when a source URL contains a scheme, and the prompt flag is determined by tag presence, an attacker can store a local filesystem path as the model version artifact location. Subsequent calls to /api/2.0/mlflow/model-versions/get-artifact then return the file contents. This results in a confidentiality breach of whatever files the MLflow server process has permission to read, potentially including sensitive configuration, credentials, or data files.

Affected Systems

MLflow by MLflow is affected. All releases from 3.10.0 up to, but not including, 3.17.0 are vulnerable. Version 3.17.0 and later contain the fix.

Risk and Exploitability

With a CVSS score of 7.5, the vulnerability presents a high impact technical weakness. The EPSS score is 2%, and it is not listed in the CISA KEV catalog, so the public exploit probability is low but nonzero. The vulnerability can be exploited by sending crafted requests to an MLflow instance that allows unauthenticated API access, or by any authenticated user in a basic‑auth configuration that can create registered models. The server‑side file read is performed by the MLflow process, so the attack requires no special privileges beyond standard server access. Without any mitigating controls, an attacker could read arbitrary files, exposing data confidentiality.

Remediation

No solution or workaround provided in the CVE record.

OpenCVE Recommended Actions

  • Upgrade MLflow to version 3.17.0 or later, which removes the path validation flaw.
  • If an upgrade cannot be performed immediately, restrict the API by requiring authentication and limiting model creation privileges to trusted users; consider disabling the prompt feature if available via configuration.
  • When possible, run the MLflow server in a container or chrooted environment and align its filesystem permissions so that only trusted directories are accessible, thereby limiting the scope of any successful file read.

Generated by OpenCVE AI on November 3, 2026 at 14:28 UTC.

Tracking

Sign in to view the affected projects.

Advisories

Source ID Title
Github GHSA Github GHSA GHSA-m65p-9j7q-6w33 MLflow vulnerable to arbitrary file read on unauthenticated servers via prompt model version source
History
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'yes', 'Exploitation': 'none', 'Technical Impact': 'partial'}, 'version': '2.0.3'}

Type Values Removed Values Added
First Time appeared Mlflow
Mlflow mlflow
Vendors & Products Mlflow
Mlflow mlflow
Type Values Removed Values Added
Description MLflow is an open source AI engineering platform for agents, large language models, and machine learning models. From 3.10.0 until 3.17.0, the prompt branch in _create_model_version() accepts schemeless relative sources because path validation runs only when a source has a URL scheme, and mlflow.prompt.is_prompt is interpreted by tag presence rather than a true value. A remote caller can store a server-local path as the model version artifact location and use /api/2.0/mlflow/model-versions/get-artifact to read files accessible to the MLflow server process. Default unauthenticated servers require no privileges, while basic-auth deployments are exposed to authenticated users who can create registered models. This issue is fixed in version 3.17.0.
Title MLflow: Arbitrary file read on unauthenticated MLflow servers via prompt model version source
Weaknesses CWE-22
References
Metrics cvssV3_1

{'score': 7.5, 'vector': 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:N/A:N'}

Weaknesses
  • CWE-22

    Improper Limitation of a Pathname to a Restricted Directory ('Path Traversal')

Sources

Status
PUBLISHED
Assigner
GitHub_M
Published
Updated
Oct 28, 2026, 4:13 PM
Reserved
Oct 26, 2026, 8:00 PM
Status
Awaiting Analysis
Published
Oct 27, 2026, 8:17 PM
Modified
Oct 28, 2026, 5:18 PM
Severity
Important
Public date
Oct 27, 2026, 7:42 PM