by Kastro
Paper/Archive/CVE-2026-114053

CVE-2026-114053

vllm-project vllm — vLLM 0.7.3 before 0.31.0 RCE via mm_processor_kwargs.code_revision

EPSS 0.5%CVSS 9.2
critical.is priorityof 100
36
CVSS 9.2 (CNA)9.2 / 10 × 3027.6Not in CISA KEV+300.0EPSS percentile 40.0.400 × 208.0No public exploit+100.0SSVC exploitation: none+100.0
Description · from the CVE record

vLLM from 0.7.3 before 0.31.0 forwards the request-supplied mm_processor_kwargs.code_revision to Transformers AutoProcessor.from_pretrained, letting clients select which revision of a model's dynamic processor code is imported when trust_remote_code is enabled. An attacker controlling a revision in the served model repository can send a multimodal request to the unauthenticated /tokenize endpoint to execute arbitrary Python in the vLLM API process.

Affected

ProductAffectedFixed inSource
vllm0.7.3 – < 0.31.00.31.0CNA

Scores by source

VulnCheck (CNA)9.2CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:H/VI:H/VA:H/SC:N/SI:N/SA:N
VulnCheck (CNA)8.1CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
CISA ADPSSVCExploitation: none · Automatable: no · Technical impact: total

References

advisorygithub.com/vllm-project/vllm/security/advisories/GHSA-h3rc-6mm3-gc2mpatchgithub.com/vllm-project/vllm/commit/570735520942b6428ae98f1eb442c6d88adae55ewriteupvulncheck.com/advisories/vllm-0.7.3-before-0.31.0-rce-via-mm-processor-kwargs-code-revision