CVE-2026-5817
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.
Does this matter?
High impact if exploited, but EPSS currently rates exploitation as unlikely (0.22%). Schedule it in the normal patch cycle and watch for a rise in EPSS or a public exploit.
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.
- CVSS 4.0
- 8.8 HIGHCVSS:4.0/AV:L/AC:L/AT:P/PR:L/UI:N/VC:H/VI:H/VA:H/SC:H/SI:H/SA:H/E:X/CR:X/IR:X/AR:X/MAV:X/MAC:X/MAT:X/MPR:X/MUI:X/MVC:X/M
- EPSS
- 0.22% probability · 13th percentile
- CISA KEV
- Not listed
- Weakness
- CWE-829
- Affected
- docker/docker desktop
- Source
- security@docker.com
References
- https://docs.docker.com/desktop/release-notes/#4680Release Notes
Source: NVD record, EPSS from FIRST.org, KEV from CISA. Refreshed daily.