CVE-2026-5817

HighCVSS 8.2

Published: 22/05/2026(UTC)Updated: 23/07/2026(UTC)Translated: 25/07/2026(UTC)NVD NIST

Exploitation Probability (EPSS)

Low risk

0.22%

13th percentile - higher than 13% of all known CVEs

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 arbitrary Python file execution from the model, enabling remote code execution on the Docker host as the Docker Desktop user.

Risk Assessment

Any container on the Docker network can trigger arbitrary code execution on the host, leading to full system compromise.

Recommendation

Restrict access to the model-runner.docker.internal API and avoid using untrusted models from OCI registries.

Other vulnerabilities in Docker Model Runner

CVE-2026-5843High
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. No sandboxing allows remote code execution on the Docker host as the Docker Desktop user.

Original NVD description (English source)

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.