CVE-2026-34760
vLLM: Downmix Implementation Differences as Attack Vectors Against Audio AI Models
- State
- Published
- Severity
- 5.9 Medium
- Published
- Apr 2, 2026
- Updated
- Apr 3, 2026
- Reserved
- Mar 30, 2026
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.
Scores
| CVSS | Score | Vector | Source |
|---|---|---|---|
| 3.1 | 5.9 Medium | CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:H/A:L | GitHub_M |
Weaknesses
- CWE-20 Improper Input Validation
Affected Products
| Product | Versions |
|---|---|
vllmvllm-project |
|
References
Every record here comes from the CVE Program’s CVE List. © 1999–2026 The MITRE Corporation. CVE is a registered trademark of The MITRE Corporation. The records are reproduced under the CVE Program terms of use.