---
id: "GHSA-f623-hjm8-5wg5"
kind: "vulnerability"
aliases: ["GHSA-f623-hjm8-5wg5"]
title: "Duplicate Advisory: vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code execution"
severity: "critical"
cvss: 8.1
epss: null
exploited: false
status: "withdrawn"
category: "ai-tooling"
products: [{"name": "vllm", "fixed": ["0.31.0"], "affected": ">= 0.7.3, < 0.31.0", "ecosystem": "pypi"}]
ecosystems: ["pypi"]
tags: ["ai-inference", "ai-tooling", "cwe-94", "python", "self-hosted", "vllm"]
not_affected_if: ["auth-required", "default-config", "managed-hosting"]
published_at: "2026-10-21T03:32:16+00:00"
first_seen_at: "2026-10-27T15:39:16+00:00"
updated_at: "2026-10-27T15:39:47+00:00"
sources: [{"source": "ghsa", "url": "https://github.com/advisories/GHSA-f623-hjm8-5wg5", "first_seen_at": "2026-10-27T15:39:16+00:00"}]
---

# Duplicate Advisory: vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code execution

vLLM forwards a client-supplied code_revision in multimodal requests to Transformers AutoProcessor, which imports dynamic processor code from that revision when trust_remote_code is enabled. An attacker who controls a revision of the served model repository can send a multimodal request to the unauthenticated /tokenize endpoint and execute arbitrary Python in the vLLM API process.

_Summary, check and fix written by a model from the sources; labelled as such._

## Affected

- pypi:vllm affected >= 0.7.3, < 0.31.0 fixed 0.31.0

## How to check

- Check the installed version with `pip show vllm`; affected range is `>=0.7.3` and `<0.31.0`.
- Check whether the server starts with `trust_remote_code` enabled (for example `--trust-remote-code`).
- Check whether the API port is reachable from untrusted networks.

## Fix

- Upgrade vLLM to `0.31.0` or later.
- Until upgraded, disable `trust_remote_code` and do not serve models from untrusted repositories.
- Restrict network access to the vLLM API port.

## Timeline

- 2026-10-27T15:39:16+00:00 ghsa https://github.com/advisories/GHSA-f623-hjm8-5wg5

## From the sources

## Duplicate Advisory

This advisory has been withdrawn because it is a duplicate of GHSA-h3rc-6mm3-gc2m. This link is maintained to preserve external references.

## Original Description
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.

---

_Free to read at a human pace. An API with keys and an MCP server are in preparation; early access: hello@infrastack.ai_