Skip to content
Fencer Vulns
medium 5.9

CVE-2026-34760 json

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

published 2026-04-02 7 mo ago modified 2026-07-24 3 mo ago also EUVD-2026-18522 , GHSA-6c4r-fmh3-7rh8 , PYSEC-2026-2299

cvss
5.9
version 3.1, CNA
cisa kev
Not listed
not in CISA's catalog
epss
0.3%
EPSS 0.3% chance of exploitation in 30 days, percentile 20%
public exploits
0
none known

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 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.

Patch / fix

  • pypi vllm 0.18.0
  • vllm-project vllm 0.18.0
  • vllm vllm 0.18.0

Vendor advisories

What the vendors themselves say about their products: affected, fixed or not affected, and what to do.

red hat product security vex CVE-2026-34760 updated cvss 5.9

vLLM: Librosa: numpy: Librosa: AI model data integrity impact due to audio processing discrepancy

  • Red Hat AI Inference Server rhaiis/vllm-cpu-rhel9, rhaiis/vllm-tpu-rhel9, rhaiis/vllm-cuda-rhel9, rhaiis/vllm-rocm-rhel9, rhaiis/vllm-spyre-rhel9, rhaiis/vllm-neuron-rhel9, +1 more Red Hat Enterprise Linux AI (RHEL AI) 3 rhelai3/bootc-cuda-rhel9, rhelai3/bootc-rocm-rhel9, rhelai3/bootc-aws-cuda-rhel9, rhelai3/bootc-gcp-cuda-rhel9, rhelai3/bootc-azure-cuda-rhel9, rhelai3/bootc-azure-rocm-rhel9 Red Hat OpenShift AI (RHOAI) rhoai/odh-vllm-cpu-rhel9, rhoai/odh-vllm-cuda-rhel9, rhoai/odh-vllm-rocm-rhel9, rhoai/odh-vllm-gaudi-rhel9, rhoai/odh-kserve-agent-rhel9, rhoai/odh-kserve-router-rhel9, +2 more
    affected: workaround, no fix available “Fix deferred”

Red Hat CSAF VEX © Red Hat, Inc., from https://security.access.redhat.com/data/csaf/v2/vex/; changed: reduced to product statuses and remediations. CC BY 4.0 · csaf

CISA SSVC

Exploitation
none
Automatable
no
Technical impact
partial
Assessed
7 mo ago

Exploit prediction

0.28% chance of exploitation in the next 30 days, higher than 20% of all scored CVEs (FIRST EPSS, 2026-11-08).

CVSS scores

assigned by version score vector
CNA 3.1 5.9 CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:H/A:L
GitHub 3.1 5.9 CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:N/I:H/A:L
NVD 3.1 7.1 CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:L
OSV 3.1 7.1 CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:H/A:L

Affected packages

ecosystem package affected fixed in
pypi vllm
>=0.5.5, <0.18.0
0.18.0

Affected products

  • vllm-project vllm

Classification

Weakness
  • CWE-20 Improper Input Validation

References

Sources

Fencer Vulns keeps one entry per vulnerability: the CVE id wins, then GHSA, and records that alias each other are merged, except that two CVEs always keep separate entries. The same data is available as JSON.

CPEs and fixed versions use Anchore's correction of NVD's data: Added CPE configurations because not yet analyzed by NVD.