CVE-2026-93840

Description

vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). Attackers can supply token IDs above the output vocabulary that pass validation, causing LogitBiasState to corrupt GPU logits state and allow concurrent requests to sample tokens outside their allowlists.

References

• https://github.com/vllm-project/vllm
• https://github.com/vllm-project/vllm/blob/v0.28.0/vllm/sampling_params.py#L881-L903
• https://github.com/vllm-project/vllm/blob/v0.28.0/vllm/v1/worker/gpu/sample/logit_bias.py#L179-L191
• https://github.com/vllm-project/vllm/commit/5b0e5b69ac1a3884a6479c9537789c95263cc804
• https://github.com/vllm-project/vllm/pull/49080