vLLM: Downmix Implementation Differences as Attack Vectors Against Audio AI...

5.9 / 10

MEDIUM

CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:H/A:L

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.

Basic Information

ID CVE-2026-34760

Source GitHub_M

Published Apr 2, 2026 at 18:59

Modified Apr 3, 2026 at 14:42

Affected Product

Vendor vllm-project

Product vllm

Version >= 0.5.5, < 0.18.0

Affected Versions vllm-project vllm >= 0.5.5, < 0.18.0

CWE Classification

CWE-20

References

{
    "lastseen": "",
    "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.",
    "published": "2026-04-02T18:59:49.638Z",
    "modified": "2026-04-03T14:42:34.842Z",
    "type": "cve",
    "title": "vLLM: Downmix Implementation Differences as Attack Vectors Against Audio AI Models",
    "source": "GitHub_M",
    "references": "https://github.com/vllm-project/vllm/security/advisories/GHSA-6c4r-fmh3-7rh8\nhttps://github.com/vllm-project/vllm/pull/37058\nhttps://github.com/vllm-project/vllm/commit/c7f98b4d0a63b32ed939e2b6dfaa8a626e9b46c4\nhttps://github.com/vllm-project/vllm/releases/tag/v0.18.0",
    "id": "CVE-2026-34760",
    "bulletinFamily": "",
    "cwe": [
        "CWE-20"
    ],
    "cvelist": null,
    "sourceData": "vllm-project vllm >= 0.5.5, < 0.18.0",
    "sourceHref": "",
    "cvss": {
        "score": 5.9,
        "severity": "MEDIUM",
        "vector": "CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:H/A:L",
        "version": "3.1"
    },
    "cvss2": [],
    "cvss3": {
        "version": "",
        "vectorString": "",
        "baseScore": 0,
        "baseSeverity": "",
        "attackVector": "",
        "attackComplexity": "",
        "privilegesRequired": "",
        "userInteraction": "",
        "scope": "",
        "confidentialityImpact": "",
        "integrityImpact": "",
        "availabilityImpact": "",
        "cvssV3": {
            "version": "",
            "vectorString": "",
            "baseScore": 0,
            "baseSeverity": "",
            "attackVector": "",
            "attackComplexity": "",
            "privilegesRequired": "",
            "userInteraction": "",
            "scope": "",
            "confidentialityImpact": "",
            "integrityImpact": "",
            "availabilityImpact": ""
        }
    },
    "href": "",
    "category_name": "CVE",
    "post_link": "",
    "product": "vllm",
    "version": ">= 0.5.5, < 0.18.0",
    "vendor": "vllm-project",
    "ai_description": "",
    "ai_severity": "",
    "ai_vendor": "",
    "ai_product": "",
    "ai_version": "",
    "ai_score": 0
}

vLLM: Downmix Implementation Differences as Attack Vectors Against Audio AI Models_CVE-2026-34760