CVE-2026-34760
CWE-20Published: April 2, 2026· Updated: Apr 3, 2026
Official 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.
Technical Analysis
CVE-2026-34760 can be exploited remotely over the network without requiring physical or adjacent access, significantly expanding the attack surface for threat actors.
Exploitation requires low privileges, which limits the exposure to scenarios where an attacker has already gained initial access.
A successful exploit results in full integrity compromise (data manipulation), with a CVSS base score of 5.9.
CVSS v3.1 Vector Breakdown
Exploit & PoC Resources
All References (4)
Quick Facts
Related CVEs (CWE-20)
Recommended Actions
- →Apply vendor patches immediately
- →Monitor CVE-2026-34760 in threat intel feeds
- →Review IDS/IPS signatures for exploitation attempts