CVE-2026-22807

HIGH

vllm 0.10.1-0.13.0 - Remote Code Execution via Hugging Face auto_map Dynamic Module Loading

Title source: llm
STIX 2.1

Exploitation Summary

EIP tracks 2 public exploits for CVE-2026-22807. PoCs published by XiaomingX, otakuliu.

AI-analyzed exploit summary This repository contains a functional proof-of-concept exploit for CVE-2026-22807, demonstrating a supply chain attack via AI model loading. The exploit leverages a TOCTOU (Time-of-Check Time-of-Use) vulnerability in AI inference frameworks, where malicious code is executed before security checks are performed.

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.

Exploits (2)

github WORKING POC 10 stars
by XiaomingX · pythonpoc
https://github.com/XiaomingX/data-cve-poc-py-v1/tree/main/2026/CVE-2026-22807

This repository contains a functional proof-of-concept exploit for CVE-2026-22807, demonstrating a supply chain attack via AI model loading. The exploit leverages a TOCTOU (Time-of-Check Time-of-Use) vulnerability in AI inference frameworks, where malicious code is executed before security checks are performed.

Classification
Working Poc 100%
Attack Type
Rce
Complexity
Moderate
Reliability
Reliable
Target: AI model loading frameworks (e.g., vLLM, Transformers)
No auth needed
Prerequisites: Python 3.x · ability to place malicious model files in a directory accessible to the target system
devstral-2 · analyzed Feb 27, 2026 Full analysis →
nomisec WORKING POC
by otakuliu · poc
https://github.com/otakuliu/CVE-2026-22807_Range

This PoC demonstrates a supply chain attack (CVE-2026-22807) exploiting a TOCTOU vulnerability in AI model loaders, where remote code execution occurs before trust checks are enforced.

Classification
Working Poc 100%
Attack Type
Rce
Complexity
Moderate
Reliability
Reliable
Target: AI model loaders (simulated as MiniLLM)
No auth needed
Prerequisites: Python 3.x · Write access to generate malicious model files
devstral-2 · analyzed Feb 16, 2026 Full analysis →

Scores

CVSS v3 8.8
EPSS 0.0003
EPSS Percentile 8.9%
Attack Vector NETWORK
CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H

CISA SSVC

Vulnrichment
Exploitation none
Automatable no
Technical Impact total

Details

CWE
CWE-94
Status published
Products (2)
pypi/vllm 0.10.1 - 0.14.0PyPI
vllm/vllm 0.10.1 - 0.14.0
Published Jan 21, 2026
Tracked Since Feb 18, 2026