Review reviewHigh

CVE-2026-34760

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.

CVSS: 7.1
EPSS: 0.27%
CISA KEV: Not listed
Published: 2026.04.03

PRIORITY ASSESSMENT

Review review

The CVSS severity warrants an early asset and exposure review.

Known exploitationNot established by KEV

Exploit probability0.27%

Technical severityCVSS 7.1

Vulnerability overview

Affected product and versions

Product: vllm-project vllm
Affected versions: >= >= 0.5.5, < 0.18.0, >= 0.5.5 < 0.18.0
Fixed versions: 0.18.0

Recommended response sequence

Confirm exposure before applying a vendor-supported change.

Full remediation guide

1
Identify
Confirm that vllm-project vllm and an affected version are present.
2
Prioritize
Combine exploitation signals with asset exposure and business criticality.
3
Remediate
Follow the vendor advisory or supported update path and preserve rollback options.
4
Verify
Recheck the version, service health, access paths, and relevant logs.

Technical data

CVSS vector: CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:L
CWE: CWE-20