---
id: CVE-2026-5817
title: >-
  The vllm-metal inference backend in Docker Model Runner on macOS
  unconditionally sets trust_remote_code=True when loading model tokenizers, and
  runs without sandboxing
summary: >-
  The vllm-metal inference backend in Docker Model Runner on macOS
  unconditionally sets trust_remote_code=True when loading model tokenizers, and
  runs without sandboxing. This causes
  transformers.AutoTokenizer.from_pretrained() to import a…
severity: high
cvss: 8.2
cvssVector: 'CVSS:3.1/AV:L/AC:L/PR:L/UI:R/S:C/C:H/I:H/A:H'
cwe:
  - CWE-829
vendor: docker
product: docker_desktop
affected:
  - 'docker_desktop >= 4.62.0, < 4.68.0'
patched:
  - docker_desktop 4.68.0
published: '2026-05-22'
updated: '2026-07-23'
source: NVD
sourceUrl: 'https://nvd.nist.gov/vuln/detail/CVE-2026-5817'
references:
  - url: 'https://docs.docker.com/desktop/release-notes/#4680'
    label: security@docker.com
tags:
  - nvd
  - exploit-available
epss: 0.00224
epssPercentile: 0.13353
ingestedAt: '2026-07-23T11:17:34.111Z'
exploits:
  github: 2
  githubRepos:
    - 'https://github.com/davidrxchester/CVE-2026-5817'
    - 'https://github.com/gouldnicholas/CVE-2026-5817-PoC'
  checkedAt: '2026-09-24T07:53:09.469Z'
exploitAvailable: true
---

## Overview

The vllm-metal inference backend in Docker Model Runner on macOS unconditionally sets trust_remote_code=True when loading model tokenizers, and runs without sandboxing. This causes transformers.AutoTokenizer.from_pretrained() to import and execute arbitrary Python files included in any model pulled from an OCI registry, resulting in arbitrary code execution on the Docker host as the Docker Desktop user when inference is triggered.

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

## Affected

- `docker_desktop >= 4.62.0, < 4.68.0`

## Remediation

Upgrade past the affected range:

- `docker_desktop 4.68.0`