CVE-2026-5817 PUBLISHED

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

Assigner: Docker
Reserved: 08.04.2026 Published: 22.05.2026 Updated: 22.05.2026

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.

Metrics

CVSS 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
CVSS Score: 8.8

Product Status

Vendor Docker
Product Docker Desktop
Versions Default: unaffected
  • affected from 4.62.0 to 4.68.0 (excl.)

Affected Configurations

Docker Model Runner enabled with the vllm-metal inference backend on macOS

Workarounds

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

Credits

  • David Rochester (@davidrxchester) finder
  • Nicholas Gould (@gouldnicholas) finder

References

Problem Types

  • CWE-829: Inclusion of Functionality from Untrusted Control Sphere CWE

Impacts

  • CAPEC-480 Escaping Virtualization