Handler on Duty: Xavier Mertens
Threat Level: green
| Published | 2026-09-19 23:17:10 |
|---|---|
| Last Modified | 2026-09-19 23:17:10 |
| AKA | CVE-2026-93989 |
| Summary | 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. |
| Access Vector | Local | Adjacent | Network |
|---|---|---|---|
| Access Complexity | Low | Medium | High |
| Authentication | None | Single | Multiple |
| Confidentiality | None | Partial | Complete |
| Integrity | None | Partial | Complete |
| Availability | None | Partial | Complete |
