by Kastro CVE-2026-22807
vllm-project vllm — vLLM affected by RCE via auto_map dynamic module loading during model initialization
Description · from the CVE record
vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.14.0, vLLM loads Hugging Face `auto_map` dynamic modules during model resolution without gating on `trust_remote_code`, allowing attacker-controlled Python code in a model repo/path to execute at server startup. An attacker who can influence the model repo/path (local directory or remote Hugging Face repo) can achieve arbitrary code execution on the vLLM host during model load. This happens before any request handling and does not require API access. Version 0.14.0 fixes the issue.
Affected
ProductAffectedFixed inSource
vllm>= 0.10.1, < 0.14.0–CNA
Red Hat AI Inference Server 3.2see advisory1772160593redhat-SADP
Red Hat AI Inference Server 3.2see advisory1772160625redhat-SADP
Red Hat AI Inference Server 3.3see advisory1782352919redhat-SADP
Red Hat AI Inference Server 3.3see advisory1782353093redhat-SADP
Red Hat AI Inference Server 3.3see advisory1782352847redhat-SADP
Red Hat OpenShift AI 2.25see advisory1776259063redhat-SADP
Red Hat OpenShift AI 2.25see advisory1772093278redhat-SADP
Red Hat OpenShift AI 2.25see advisory1783998774redhat-SADP
Red Hat OpenShift AI 2.25see advisory1783998857redhat-SADP
Red Hat OpenShift AI 3.3see advisory1770956034redhat-SADP
Red Hat OpenShift AI 3.4see advisory1772204861redhat-SADP
Scores by source
NVD9.8CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H
GitHub_M (CNA)8.8CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
redhat-SADP (CNA)8.8CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
CISA ADPSSVCExploitation: none · Automatable: no · Technical impact: total