CVE WatchJul 15, 2026

CVE-2026-22807

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

CVSS

8.8

High

EPSS

0.7%

p50

KEV

—

Exploit Today

0-100

Published: Jan 21, 2026 · Last modified: Jul 15, 2026 · CWE-94

EPSS · 30d

0.7%EPSS · 30 days0.7%

2026-06-302026-07-17

Technical description

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.

Official references

Related CVEs

CVECVSSEPSSKEVRExploitTitleMod.

CVE-2025-32489.8 CRI

100.0%

KEV—80Langflow Missing Authentication Vulnerability3d CVE-2026-341978.8 HIG

99.9%

KEV—80Apache ActiveMQ Improper Input Validation Vulnerability3d CVE-2026-154107.2 HIG

71.1%

KEV—71SonicWall SMA1000 Appliances Code Injection Vulnerability2d CVE-2025-670389.8 CRI

55.3%

KEV—67Lantronix EDS5000 Code Injection Vulnerability11d CVE-2021-416539.8 CRI

99.5%

——30The PING function on the TP-Link TL-WR840N EU v5 router with firmware through TL-WR840N(EU)_V5_171211 is vulnerable to remote code execution via a crafted payload in an IP address input field.9d CVE-2023-362558.8 HIG

99.0%

——30An issue in Eramba Limited Eramba Enterprise and Community edition v.3.19.1 allows a remote attacker to execute arbitrary code via the path parameter in the URL.9d