CVE-2026-93840

Low

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.

SeverityLow

CVSS3.7

Exploit probability—

ProductVllm

PublishedSep 18, 2026

First seenSep 18, 2026

Weakness typeCWE-129CWE-129

How to fix

Update

vllm vllm0.29.0 or later
vllm0.29.0 or later

Patches

github.com/vllm-project/vllm/pull/49080
github.com/vllm-project/vllm/commit/5b0e5b69ac1a3884a6479c9537789c95263cc804

Affected products

Product	Affected versions	Fixed in
vllm vllm	v < 0.29.0	0.29.0
vllmPyPI	v < 0.29.0	0.29.0

Check your version

Choose the product and enter the version you run to see whether it falls in the affected range.

In the Intelligence 360 plan

Attack Surface Management

See the assets tied to your domains, the technologies they run and known vulnerabilities in one list.