---
id: "CVE-2026-116178"
kind: "vulnerability"
aliases: ["CVE-2026-116178"]
title: "vLLM: crafted multimodal requests crash the inference engine remotely"
severity: "high"
cvss: 8.2
epss: null
exploited: false
status: "open"
category: "ai-tooling"
products: []
ecosystems: []
tags: ["ai-tooling", "inference-server", "multimodal", "python", "self-hosted", "vllm"]
not_affected_if: ["auth-required", "default-config"]
published_at: "2026-10-25T14:16:57+00:00"
first_seen_at: "2026-10-25T14:41:23+00:00"
updated_at: "2026-10-25T14:42:23+00:00"
sources: [{"source": "nvd", "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-116178", "first_seen_at": "2026-10-25T14:41:23+00:00"}]
---

# vLLM: crafted multimodal requests crash the inference engine remotely

vLLM versions from 0.20.0 before 0.30.0 let a remote caller send inconsistent multimodal metadata to the disaggregated /inference/v1/generate endpoint. The malformed request makes the engine build a negative text span in Qwen2.5-VL position handling, which terminates the engine and denies service to everyone using it.

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

## How to check

- Check the installed version with `pip show vllm` and treat anything from `0.20.0` up to but not including `0.30.0` as exposed.
- Check whether the server exposes the disaggregated `/inference/v1/generate` endpoint to untrusted clients.
- Check whether the deployment serves Qwen2.5-VL or other multimodal models.

## Fix

- Upgrade vLLM to `0.30.0` or later.
- Until upgraded, block or restrict access to `/inference/v1/generate` from untrusted networks.

## Timeline

- 2026-10-25T14:41:23+00:00 nvd https://nvd.nist.gov/vuln/detail/CVE-2026-116178

## From the sources

vLLM from 0.20.0 before 0.30.0 contains a denial of service vulnerability that allows remote attackers to crash the engine by sending inconsistent multimodal metadata to the disaggregated /inference/v1/generate endpoint. Attackers can replay a valid multimodal item with repeated or overlapping mm_placeholders ranges or mismatched kwargs_data and mm_hashes lists, producing a negative text span in Qwen2.5-VL position construction that terminates the engine.

---

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