CVE-2026-5843

Docker Model Runner container-to-host code execution via MLX-LM model_file importlib loading

Published May 22, 2026 CVSS 4.0 · 8.8 High EPSS < 1% KEV · Not listed
Description
The MLX inference backend in Docker Model Runner on macOS uses the MLX-LM library, which unconditionally imports and executes arbitrary Python files from model directories via the model_file configuration field in config.json. When a model's config.json specifies a model_file pointing to a Python file, MLX-LM uses importlib to load and execute it with no trust_remote_code gate or equivalent safety check. The MLX backend runs without sandboxing, resulting in arbitrary code execution on the Docker host as the Docker Desktop user.

Any container on the Docker network can trigger this by calling the model-runner.docker.internal API to pull a malicious model from an attacker-controlled OCI registry and request inference.

Analysis

AI-generated

Technical impact

The MLX inference backend in Docker Model Runner on macOS imports and executes arbitrary Python files listed in a model’s configuration, without any sandboxing or trust guard. Because the loader uses importlib unconditionally, a malicious model can contain code that runs as the Docker Desktop user, giving an attacker full control of the host machine's file system and resources.

Affected Systems

Docker Desktop on macOS with the Model Runner feature enabled is affected. Any container that accesses the Model Runner API can trigger the vulnerability by loading a model that points to a malicious Python file.

Risk and Exploitability

The vulnerability has a CVSS score of 8.8 and is not currently listed in the CISA KEV catalog. Its exploitability relies on an attacker-controlled OCI registry from which a container can pull a malicious model and then activate inference via the model‑runner.docker.internal API. Any container on the Docker network can initiate this, so the risk is significant for environments that run untrusted containers with Model Runner enabled.

Remediation

Vendor Workaround

Disable Docker Model Runner or only run trusted containers on Docker Desktop instances where Model Runner is enabled.


OpenCVE Recommended Actions

  • Disable Docker Model Runner or ensure it runs only trusted containers on instances where the feature is enabled.
  • Upgrade Docker Desktop to a version that includes the fix released after version 47.10, as detailed in Docker’s release notes.
  • If a patch is unavailable, block external OCI registries or restrict network access to the model‑runner.docker.internal endpoint to prevent malicious models from being retrieved.

Generated by OpenCVE AI on May 22, 2026 at 21:22 UTC.

Tracking

Sign in to view the affected projects.

Advisories

No advisories available
History
Type Values Removed Values Added
First Time appeared Apple
Apple macos
CPEs cpe:2.3:a:docker:docker_desktop:*:*:*:*:*:*:*:*
cpe:2.3:o:apple:macos:-:*:*:*:*:*:*:*
Vendors & Products Apple
Apple macos
Type Values Removed Values Added
Metrics ssvc

{'options': {'Automatable': 'no', 'Exploitation': 'none', 'Technical Impact': 'total'}, 'version': '2.0.3'}

Type Values Removed Values Added
First Time appeared Docker
Docker docker Desktop
Vendors & Products Docker
Docker docker Desktop
Type Values Removed Values Added
Description The MLX inference backend in Docker Model Runner on macOS uses the MLX-LM library, which unconditionally imports and executes arbitrary Python files from model directories via the model_file configuration field in config.json. When a model's config.json specifies a model_file pointing to a Python file, MLX-LM uses importlib to load and execute it with no trust_remote_code gate or equivalent safety check. The MLX backend runs without sandboxing, resulting in arbitrary code execution on the Docker host as the Docker Desktop user. Any container on the Docker network can trigger this by calling the model-runner.docker.internal API to pull a malicious model from an attacker-controlled OCI registry and request inference.
Title Docker Model Runner container-to-host code execution via MLX-LM model_file importlib loading
Weaknesses CWE-829
References
Metrics cvssV3_1

{'score': 8.2, 'vector': 'CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:C/C:H/I:H/A:H'}

cvssV4_0

{'score': 8.8, 'vector': 'CVSS:4.0/AV:L/AC:L/AT:P/PR:L/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H'}

Weaknesses
  • CWE-829

    Inclusion of Functionality from Untrusted Control Sphere

Sources

Status
PUBLISHED
Assigner
Docker
Published
Updated
May 27, 2026, 3:55 AM
Reserved
Apr 8, 2026, 5:43 PM
Status
Analyzed
Published
May 22, 2026, 8:16 PM
Modified
Jul 23, 2026, 11:10 AM