Technical impact
vLLM versions prior to validate the sample rate in FLAC audio headers used by the transcription endpoint. This missing check allows an authenticated user to craft a header with an inflated sample rate, causing the server to allocate an excessive amount of memory during transcription. The resulting memory exhaustion triggers a crash of the API server, leading to a denial of service that impacts all tenants that rely on the service.
Affected Systems
Any deployment of the open‑source library vLLM from vllm-project using a 0.28.0‑older release is susceptible to this flaw, as identified in the advisory.
Risk and Exploitability
The CVSS score of 7.1 indicates a high impact. The EPSS score of < 1% indicates a very low probability of exploitation, but the vulnerability is not listed in the CISA KEV catalog. Explo a specially crafted FLAC file. Successful service interruption for all tenants. Attackers with valid credentials can therefore achieve widespread denial of service with minimal effort once an account is compromised.