Skip to content
OpenAdvisories
High · 8.2 Public exploit CVE-2026-5817 EUVD-2026-31493GHSA-cgqp-ww2v-6rjh

Docker Model Runner container-to-host code execution via unsandboxed trust_remote_code in Python inference backends

Published May 22, 2026 · Updated May 27, 2026 · Assigned by Docker

Vendor: Docker Product: Docker Desktop

Description

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 products

VendorProductAffectedFixed
Docker Docker Desktop 4.62.0 – < 4.68.0 4.68.0

Public exploits

Exploit or proof-of-concept code for this vulnerability is publicly available, which lowers the bar for attackers. Treat it as a reason to patch sooner.

Third-party code, not reviewed by us. Public PoCs are sometimes fake or contain malware. Read the code first and run it only in an isolated lab, against systems you are authorised to test.

GitHub 2 repositories

  • gouldnicholas/CVE-2026-5817-PoC ★ 0 · May 22, 2026

    Docker Model Runner container-to-host RCE / Escape: A critical vulnerability that allows for container-to-host code execution in the Docker Model Runner MLX / SGLANG / VLLM inference backend.

  • davidrxchester/CVE-2026-5817 ★ 0 · May 20, 2026

    Docker Container-to-Host Remote Code Execution POC via vllm-metal trust_remote_code=True

Sources: PoC-in-GitHub, Metasploit Framework, Exploit-DB — updated daily. Only links are listed; no exploit code is hosted here.

Advisories 1

References