PYSEC-2026-4000
At a glance
- Published
- Updated
- CVE
- CVE-2026-93989
- CVSS_V3
- CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:L/A:N
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 packages
| Package | Versions |
|---|---|
| PyPI · vllm | introduced 0, fixed 0.30.0 |
References
- https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/sampling_params.py#L694-L753 WEB
- https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/v1/worker/gpu/sample/bad_words.py WEB
- https://www.vulncheck.com/advisories/vllm-through-0.29.0-cross-request-logits-corruption-via-bad-words ADVISORY
- https://github.com/vllm-project/vllm/pull/48824 FIX
- https://github.com/vllm-project/vllm PACKAGE
Sources
| Source | Licence |
|---|---|
| OSV.dev | CC-BY 4.0 — PyPA Advisory Database |