| Component | vLLM |
| Severity | medium (CVSS 6.5) |
| CVSS vector | CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:U/C:N/I:N/A:H |
| Published (NVD) | 2026-09-26 |
| Last modified | 2026-10-07 |
| Affected versions | vllm < 0.29.0 |
| Fixed in | 0.29.0 |
| Weakness | CWE-129 |
| CISA KEV | not listed |
NVD description
vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing operation (index_put_) and triggers a device-side assertion. An authenticated API user can send a single malformed completion request that returns 500 Internal Server Error and puts EngineCore into a fatal state, causing subsequent requests to fail until the service is restarted (denial of service).
What to do
- Upgrade vLLM to
0.29.0or later. - Do not expose the component's API or UI to the internet without authentication in front of it.
Sources
- NVD: CVE-2026-100654
- https://github.com/vllm-project/vllm/security/advisories/GHSA-v5gm-qgmv-gc6c
- https://www.vulncheck.com/advisories/vllm-before-0.29.0-denial-of-service-via-out-of-range-stop-token-ids
Read more
Other vLLM CVEs
vLLM before 0.29.0 passes uploaded audio to the PyAV/FFmpeg av.open() fallback with only a file-size check, exposing FFmpeg's IAMF parser he...
A security flaw has been discovered in vllm-project vLLM up to 0.31.0. This impacts the function get_token_bin_counts_and_mask of the file v...
A security vulnerability has been detected in vllm-project vLLM up to 0.31.0. This impacts the function conv_ssm_forward of the file vllm/mo...
vLLM is an inference and serving engine for large language models. Prior to 0.30.0, flash late-interaction scoring at the /score and /rerank...
vLLM is an inference and serving engine for large language models. Prior to 0.30.0, Harmony tool continuations submitted through "POST /v1/r...
vLLM is an inference and serving engine for large language models. Prior to 0.28.0, the default mirrored multimodal LRU cache can commit a m...