vLLM's audio decode-duration guard is only applied to the speech-to-text endpoint but not the chat endpoint, allowing unauthenticated clients to submit compressed audio files that expand to multiple GB and cause out-of-memory denial of service attacks.
Disclosed 16 September 2026 · Record updated 17 September 2026
Unauthenticated remote denial of service via memory amplification on audio-capable model deployments with default no-auth configuration.
VulnerabilitiesA missing size guard on the chat endpoint lets a few kilobytes of compressed audio expand into gigabytes of memory, according to a GitHub advisory published on 16 September 2026.
17 Sept 2026
