CVE-2026-115916GHSA-q97c-p368-75c4
vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS)
vllm-project vllm
CVSS
6.5MediumExploitationNo known exploitationNot in CISA KEV, which does not prove it is unexploitedCISA KEV
Exploit odds (EPSS)No scoreNo FIRST EPSS score in this snapshot yet
Fixed inNone listedNo source names a fixed version. Check the vendor advisory below
What fixes it
No source in this snapshot names a fixed version. The vendor advisories under References are the place to check.
Description
vLLM is an inference and serving engine for large language models. Prior to 0.32.0, vllm/tokenizers/hf.py computes max_token_id with an off-by-one upper bound, allowing a pre-tokenized prompt ID equal to the vocabulary size to pass validation. For a tensor-parallel size of one and an exact multiple-of-64 vocabulary, VocabParallelEmbedding.forward does not mask that value before F.embedding uses it, causing a CUDA device-side assertion and terminating the engine. This issue is fixed in version 0.32.0.
Text from CVE List
Affected versions 1
- vllmaffected: < 0.32.0CVE List
References 4
- github.com/vllm-project/vllm/commit/37cecc8141e2dcba9dd534897a164d4baaf27229Fixvia CVE List
- github.com/vllm-project/vllm/releases/tag/v0.32.0Fixvia CVE List
- github.com/vllm-project/vllm/security/advisories/GHSA-q97c-p368-75c4Advisoryvia CVE List
- github.com/vllm-project/vllm/pull/59491Referencevia CVE List