Description
vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). Attackers can supply out-of-bounds token indices that corrupt logits memory of concurrent requests, causing different in-flight HTTP requests to return incorrect tokens.
Affected products
Version statuses are supplied by the assigning organization. Consult the original record for detailed ranges and platform conditions.
vllm-project vllm
Default version status: unaffected
0 to ≤ 0.29.0affected
pkg:pypi/vllmReferences
- https://github.com/vllm-project/vllm/pull/48824 ↗
- https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/v1/worker/gpu/sample/bad_words.py ↗
- https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/sampling_params.py#L694-L753 ↗
- https://github.com/vllm-project/vllm ↗
- https://www.vulncheck.com/advisories/vllm-through-0.29.0-cross-request-logits-corruption-via-bad-words ↗