---
id: CVE-2026-57173
title: vLLM is an inference and serving engine for large language models
summary: >-
  vLLM is an inference and serving engine for large language models. Prior to
  0.24.0, the input_audio handling path for /v1/chat/completions calls
  AudioMediaIO.load_bytes or AudioMediaIO.load_file without passing
  VLLM_MAX_AUDIO_DECODE_DURA…
severity: medium
cvss: 6.5
cvssVector: 'CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H'
cwe:
  - CWE-770
vendor: vllm-project
product: vllm
affected:
  - vllm < 0.24.0
patched:
  - vllm 0.24.0
published: '2026-09-16'
updated: '2026-09-18'
sourceUpdated: '2026-09-18T18:17:07.480'
source: NVD
sourceUrl: 'https://nvd.nist.gov/vuln/detail/CVE-2026-57173'
references:
  - url: >-
      https://github.com/vllm-project/vllm/commit/3d20275bb4d434f53055c3c0b645fd8bb072965e
    label: security-advisories@github.com
  - url: 'https://github.com/vllm-project/vllm/pull/45908'
    label: security-advisories@github.com
  - url: 'https://github.com/vllm-project/vllm/releases/tag/v0.24.0'
    label: security-advisories@github.com
  - url: >-
      https://github.com/vllm-project/vllm/security/advisories/GHSA-hcwq-8wjf-3gcr
    label: security-advisories@github.com
  - url: 'https://nvd.nist.gov/vuln/detail/CVE-2026-57173'
  - url: 'https://github.com/advisories/GHSA-hcwq-8wjf-3gcr'
  - url: 'https://github.com/vllm-project/vllm'
  - url: >-
      https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-57173.json
  - url: 'https://access.redhat.com/security/cve/CVE-2026-57173'
  - url: 'https://bugzilla.redhat.com/show_bug.cgi?id=2535557'
  - url: 'https://www.cve.org/CVERecord?id=CVE-2026-57173'
tags:
  - nvd
  - cve.org
  - ghsa
  - pip
  - osv
  - csaf
  - vex
  - red-hat
ssvc:
  exploitation: none
  automatable: 'no'
  technicalImpact: partial
  timestamp: '2026-09-18T18:11:39.227201Z'
epss: 0.00658
epssPercentile: 0.4971
aliases:
  - GHSA-hcwq-8wjf-3gcr
ecosystem: pip
ingestedAt: '2026-09-16T16:59:56.552Z'
---

## Overview

vLLM is an inference and serving engine for large language models. Prior to 0.24.0, the input_audio handling path for /v1/chat/completions calls AudioMediaIO.load_bytes or AudioMediaIO.load_file without passing VLLM_MAX_AUDIO_DECODE_DURATION_S to the shared audio decoder. An unauthenticated client can therefore submit a small compressed audio input that expands into a very large float32 PCM allocation, bypassing the duration guard already used by /v1/audio/transcriptions and causing an out-of-memory worker crash. Inline data URLs reach this path without being bounded by VLLM_AUDIO_FETCH_TIMEOUT. The issue affects deployments serving an audio-capable model, and authentication changes only the deployment-specific reachability. This issue is fixed in version 0.24.0.

## Remediation

Refer to the linked advisories for vendor-supplied fixes and affected version ranges.

## Package advisory (CVE-2026-57173)

Affected packages:

- `vllm <= 0.23.0`

Patched in:

- `vllm 0.24.0`

Source: https://github.com/advisories/GHSA-hcwq-8wjf-3gcr

## Vendor advisories

- **Red Hat VEX** · Moderate · affected: Red Hat AI Inference Server, Red Hat Enterprise Linux AI (RHEL AI) 3, Red Hat OpenShift AI (RHOAI) · no fix planned: Red Hat AI Inference Server, Red Hat Enterprise Linux AI (RHEL AI) 3, Red Hat OpenShift AI (RHOAI) · updated 2026-09-16 · [vex](https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-57173.json)