Improper Validation of Array Index Affecting vllm package, versions [0.22.0,0.24.0)
Threat Intelligence
Snyk has a proof-of-concept or detailed explanation of how to exploit this vulnerability.
The probability is the direct output of the EPSS model, and conveys an overall sense of the threat of exploitation in the wild. The percentile measures the EPSS probability relative to all known EPSS scores. Note: This data is updated daily, relying on the latest available EPSS model version. Check out the EPSS documentation for more details.
Do your applications use this vulnerable package?
In a few clicks we can analyze your entire application and see what components are vulnerable in your application, and suggest you quick fixes.
Test your applications- Snyk IDSNYK-PYTHON-VLLM-20158366
- published27 Sept 2026
- disclosed26 Sept 2026
- creditUnknown
Introduced: 26 Sep 2026
NewCVE-2026-100652 (opens in a new tab)How to fix?
Upgrade vllm to version 0.24.0 or higher.
Overview
vllm is an A high-throughput and memory-efficient inference and serving engine for LLMs
Affected versions of this package are vulnerable to Improper Validation of Array Index via missing vocabulary-bound validation of stop_token_ids in the Rust HTTP and gRPC frontends. When a request is submitted with min_tokens greater than zero and a stop_token_ids entry outside the range [0, vocab_size), the out-of-vocabulary token ID is forwarded through EngineCoreSamplingParams.all_stop_token_ids into MinTokensLogitsProcessor, where it is used as a CUDA tensor index in logits.index_put_. This triggers a device-side assertion that leaves EngineCore in a fatal state requiring a full service restart. The Rust frontends already validate neighboring token-ID fields such as prompt, allowed_token_ids, and logit_bias against vocabulary bounds, but apply no equivalent check to stop_token_ids before lowering the request into engine-facing sampling state.