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.
Metrics
CVSS 3.1 4.3/10
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:NWeakness enumeration
Affected software
| Vendor | Product | Versions |
|---|---|---|
| vllm-project | vllm | <= 0.29.0 |
References
- https://github.com/vllm-project/vllmProduct
- https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/sampling_params.py#L694-L753Product
- https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/v1/worker/gpu/sample/bad_words.pyProduct
- https://github.com/vllm-project/vllm/pull/48824Issue Tracking, Patch
- https://www.vulncheck.com/advisories/vllm-through-0.29.0-cross-request-logits-corruption-via-bad-wordsThird Party Advisory
Timeline
- Published
- Sep 19, 2026
- Last modified
- Sep 28, 2026
- NVD status
- Analyzed