---
id: CVE-2026-34760
aliases:
  - PYSEC-2026-2299
  - GHSA-6c4r-fmh3-7rh8
title: >-
  vLLM is an inference and serving engine for large language models (LLMs). From
  version 0.5.5 to before version 0.18.0, Librosa defaults t…
summary: >-
  vLLM is an inference and serving engine for large language models (LLMs). From
  version 0.5.5 to before version 0.18.0, Librosa defaults to using numpy.mean
  for mono downmixing (to_mono), while the international standard ITU-R BS.775-4
  sp…
severity: high
cvss: 7.1
cvssVector: 'CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:L'
vendor: vllm
product: vllm
ecosystem: pip
affected:
  - 'vllm >= 0.5.5, < 0.18.0'
patched:
  - vllm 0.18.0
published: '2026-04-02'
updated: '2026-07-13'
source: OSV
sourceUrl: 'https://osv.dev/vulnerability/PYSEC-2026-2299'
references:
  - url: 'https://github.com/vllm-project/vllm/releases/tag/v0.18.0'
  - url: >-
      https://github.com/vllm-project/vllm/security/advisories/GHSA-6c4r-fmh3-7rh8
  - url: 'https://github.com/vllm-project/vllm/pull/37058'
  - url: >-
      https://github.com/vllm-project/vllm/commit/c7f98b4d0a63b32ed939e2b6dfaa8a626e9b46c4
tags:
  - osv
  - pip
epss: 0.00476
epssPercentile: 0.3846
ingestedAt: '2026-07-13T18:58:08.933Z'
---

## Overview

vLLM is an inference and serving engine for large language models (LLMs). From version 0.5.5 to before version 0.18.0, Librosa defaults to using numpy.mean for mono downmixing (to_mono), while the international standard ITU-R BS.775-4 specifies a weighted downmixing algorithm. This discrepancy results in inconsistency between audio heard by humans (e.g., through headphones/regular speakers) and audio processed by AI models (Which infra via Librosa, such as vllm, transformer). This issue has been patched in version 0.18.0.

## Affected packages

- `vllm >= 0.5.5, < 0.18.0`

## Remediation

Upgrade to a patched release:

- `vllm 0.18.0`