gcdRadar
archive/Patch now/AIR-2026-0437
closedclosed 21 Jan 2026Patch now2 Disclosed vulnerabilityAI

CVE-2026-22807: vLLM RCE via auto_map Dynamic Module Loading

vLLM

vLLM loads Hugging Face auto_map dynamic modules without gating on trust_remote_code, allowing arbitrary code execution at server startup. The vulnerability can be exploited by influencing the model repository path.

disclosed CVENo confirmed exploitation, so this record is tracked as a disclosed vulnerability and not counted as an incident. It moves into the counts if CISA or the vendor confirms exploitation. Why

CVSS
8.8
EPSS
0.01 · top 38% · 2026-11-13
fix status
Fix status unknown
first report
21 Jan 2026
last update
21 Jan 2026

Timeline · 2

  1. 21 Jan 2026CVE-2026-22807: vLLM affected by RCE via auto_map dynamic module loading during model initialization
  2. 21 Jan 2026CVE-2026-22807: vLLM is an inference and serving engine for large language models (LLMs).

Sources · 2

Advisories and databases

  1. 2026-01-21
    CVE-2026-22807: vLLM affected by RCE via auto_map dynamic module loading during model initialization
    GitHub Advisory Database · github.com/advisories/GHSA-2pc9-4j83-qjmr
  2. 2026-01-21

Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.

Classification

AI role
Attack on AI
categories
Infrastructure vulnerability
severity
2 · Disclosed vulnerability
fix status
Fix status unknown
targets
Framework / tooling
owasp llm top 10
LLM03:2025 Supply Chain
mitre atlas
no technique for these categories
tags
rce
dates
happened 21 Jan 2026 · first reported 21 Jan 2026
origin
GitHub Advisory Database
Vulnerability record from NVD / the GitHub Advisory Database. This product uses the NVD API but is not endorsed or certified by the NVD.
record
confidence 0.95 · created 2026-10-18 07:00Z · updated 2026-10-18 10:04Z

Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned

Copy a brief

This record as a short Markdown brief, for a ticket, a chat or a coding agent.

the brief as text
# CVE-2026-22807: vLLM RCE via auto_map Dynamic Module Loading

https://gcdradar.vercel.app/incidents/cve-2026-22807-vllm-rce-via-auto-map-dynamic-module-loading · AIR-2026-0437 · closed 2026-01-21

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-01-21
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2026-22807
- CVSS: 8.8

## Summary

vLLM loads Hugging Face auto_map dynamic modules without gating on trust_remote_code, allowing arbitrary code execution at server startup. The vulnerability can be exploited by influencing the model repository path.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-01-21: CVE-2026-22807: vLLM is an inference and serving engine for large language models (LLMs). (https://nvd.nist.gov/vuln/detail/CVE-2026-22807)
- 2026-01-21: CVE-2026-22807: vLLM affected by RCE via auto_map dynamic module loading during model initialization (https://github.com/advisories/GHSA-2pc9-4j83-qjmr)

## Sources

- 2026-01-21 GitHub Advisory Database: CVE-2026-22807: vLLM affected by RCE via auto_map dynamic module loading during model initialization https://github.com/advisories/GHSA-2pc9-4j83-qjmr
- 2026-01-21 NVD: CVE-2026-22807: vLLM is an inference and serving engine for large language models (LLMs). https://nvd.nist.gov/vuln/detail/CVE-2026-22807

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.

Related

  1. closed9.8
    vLLM CVE-2026-22778: vLLM remote code execution via video processing

    Patch now · Disclosed · since 2 Feb 2026 · CVE-2026-22778

    AI
  2. closed7.1
    vLLM CVE-2026-24779: vLLM MediaConnector vulnerable to Server-Side Request Forgery

    Patch now · Disclosed · since 28 Jan 2026 · updated 28 Jan 2026 · CVE-2026-24779

    AI
  3. closed6.5
    vLLM CVE-2026-22773: vLLM Denial of Service via Idefics3 Vision Model Payload

    Patch now · Disclosed · since 13 Jan 2026 · updated 13 Jan 2026 · CVE-2026-22773

    AI
  4. closed8.8
    vLLM CVE-2026-56340: vLLM introduces enhanced protection for sparse tensor validation DoS

    Patch now · Disclosed · since 8 Jan 2026 · updated 20 Jun 2026 · CVE-2026-56340

    AI
  5. closed7.1
    vLLM vLLM denial of service via prompt embeds on M-RoPE models

    Patch now · Disclosed · since 20 Jul 2026 · updated 20 Jul 2026 · CVE-2026-55514

    AI
  6. closed8.7
    vLLM vLLM ReDoS vulnerability in structured outputs regex compilation

    Patch now · Disclosed · since 17 Jul 2026 · updated 17 Jul 2026 · CVE-2026-55574

    AI