vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset without bounds checking against vocabulary size. Attackers can submit multimodal audio requests with tokens equal to vocabulary size, causing out-of-bounds writes that corrupt concurrent requests' sampler state and alter repetition penalty behavior.
https://github.com/vllm-project/vllm/pull/49081
https://github.com/vllm-project/vllm/blob/v0.29.0/vllm/v1/worker/gpu/sample/penalties.py#L33-L37
https://github.com/vllm-project/vllm/blob/v0.29.0/vllm/v1/worker/gpu/sample/penalties.py#L241-L255
Published: 2026-09-18
Updated: 2026-09-18
Base Score: 2.6
Vector: CVSS2#AV:N/AC:H/Au:N/C:N/I:P/A:N
Severity: Low
Base Score: 3.7
Vector: CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:L/A:N
Severity: Low
Base Score: 6.3
Vector: CVSS:4.0/AV:N/AC:L/AT:P/PR:N/UI:N/VC:N/VI:L/VA:N/SC:N/SI:N/SA:N
Severity: Medium
EPSS: 0.00239