Advisory details:
pypa / vllm/PYSEC-2026-4000
| Advisory ID | pypa/vllm/PYSEC-2026-4000 |
| Aliases |
CVE-2026-93989
PYSEC-2026-4000 |
| 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. |
| Status | Published |
| Exploitability | 0.5 |
| Weighted Severity | 3.9 |
| Risk | 1.9 |
| Affected and Fixed Packages | Package Details |
| Introduced and Fixed Package Commit Patches | Package Commit Patches Details |
| There are no known CWE. |
| System | Score | Found at |
|---|---|---|
| cvssv3.1 | 4.3 | https://github.com/pypa/advisory-database/blob/main/vulns/vllm/PYSEC-2026-4000.yaml |
| Reference id | Reference type | URL |
|---|---|---|
| https://github.com/vllm-project/vllm | ||
| 48824 | Bug | https://github.com/vllm-project/vllm/pull/48824 |
| bad_words.py | Other | https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/v1/worker/gpu/sample/bad_words.py |
| sampling_params.py#L694-L753 | Other | https://github.com/vllm-project/vllm/blob/98dff2a81d747d1dba01a47f939f48c3526d4206/vllm/sampling_params.py#L694-L753 |
| vllm-through-0.29.0-cross-request-logits-corruption-via-bad-words | Advisory | https://www.vulncheck.com/advisories/vllm-through-0.29.0-cross-request-logits-corruption-via-bad-words |
No EPSS data available for this advisory.
| Patch URL | |
|---|---|
| There are no known patches. | |
| Attack Vector (AV) | Attack Complexity (AC) | Privileges Required (PR) | User Interaction (UI) | Scope (S) | Confidentiality Impact (C) | Integrity Impact (I) | Availability Impact (A) |
|---|---|---|---|---|---|---|---|
network adjacent_network local physical |
low high |
none low high |
none required |
unchanged changed |
high low none |
high low none |
high low none |
There are no SSVC decisions available.