CVE-2026-34760: CWE-20: Improper Input Validation in vllm-project vllm
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.
AI Analysis
Technical Summary
CVE-2026-34760 affects the vLLM inference engine for large language models in versions 0.5.5 through before 0.18.0. The vulnerability arises because Librosa, used by vLLM for audio processing, defaults to numpy.mean for converting stereo audio to mono, which deviates from the ITU-R BS.775-4 standard that specifies a weighted downmixing algorithm. This improper input validation leads to inconsistencies between the audio humans hear and the audio input processed by AI models, potentially impacting the integrity of AI inference results. The issue is fixed in vLLM version 0.18.0.
Potential Impact
The vulnerability does not compromise confidentiality but can cause incorrect audio processing by AI models using vLLM, potentially leading to integrity issues in model inference outputs. Availability impact is rated low. There are no reports of active exploitation in the wild.
Mitigation Recommendations
A fix is available in vLLM version 0.18.0. Users should upgrade to this version or later to resolve the improper input validation issue related to audio downmixing. No additional mitigation steps are indicated.
CVE-2026-34760: CWE-20: Improper Input Validation in vllm-project vllm
Description
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.
AI-Powered Analysis
Machine-generated threat intelligence
Technical Analysis
CVE-2026-34760 affects the vLLM inference engine for large language models in versions 0.5.5 through before 0.18.0. The vulnerability arises because Librosa, used by vLLM for audio processing, defaults to numpy.mean for converting stereo audio to mono, which deviates from the ITU-R BS.775-4 standard that specifies a weighted downmixing algorithm. This improper input validation leads to inconsistencies between the audio humans hear and the audio input processed by AI models, potentially impacting the integrity of AI inference results. The issue is fixed in vLLM version 0.18.0.
Potential Impact
The vulnerability does not compromise confidentiality but can cause incorrect audio processing by AI models using vLLM, potentially leading to integrity issues in model inference outputs. Availability impact is rated low. There are no reports of active exploitation in the wild.
Mitigation Recommendations
A fix is available in vLLM version 0.18.0. Users should upgrade to this version or later to resolve the improper input validation issue related to audio downmixing. No additional mitigation steps are indicated.
Technical Details
- Data Version
- 5.2
- Assigner Short Name
- GitHub_M
- Date Reserved
- 2026-03-30T19:17:10.225Z
- Cvss Version
- 3.1
- State
- PUBLISHED
Threat ID: 69cec5aae6bfc5ba1dfbd810
Added to database: 4/2/2026, 7:38:18 PM
Last enriched: 4/9/2026, 10:47:15 PM
Last updated: 5/24/2026, 10:50:28 AM
Views: 212
Community Reviews
0 reviewsCrowdsource mitigation strategies, share intel context, and vote on the most helpful responses. Sign in to add your voice and help keep defenders ahead.
Want to contribute mitigation steps or threat intel context? Sign in or create an account to join the community discussion.
Actions
Updates to AI analysis require Pro Console access. Upgrade inside Console → Billing.
Need more coverage?
Upgrade to Pro Console for AI refresh and higher limits.
For incident response and remediation, OffSeq services can help resolve threats faster.
Latest Threats
Check if your credentials are on the dark web
Instant breach scanning across billions of leaked records. Free tier available.