datAInsights CVE Current

CVE-2026-93840 — vLLM before 0.29.0 Cross-Request Logits Corruption via allowed_token_ids

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 suppl…

Published: 2026-09-18 · CVSS: 6.3 · CVSS 4.0

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.

Original CVE JSON · STIX 2.1

Source: CVE Program · Revision 8b16575c39f1e1810c5e863010da073d2d3285e7