vllm vulnerability

vLLM before 0.29.0 contains a resource-limit bypass vulnerability in PyNvVideoCodec decoder allocation where sampler subclass shadowing allows independent counter increments. Unauthenticated attackers can select different sampler subclasses in video requests to exceed configured decoder limits and exhaust unaccounted GPU memory.

Published 26 Sep 2026Updated 26 Sep 20262 sources
CVSS 6.3

Record history

Record created from the first normalized source observation.
Metadata and source references refreshed.