Staging Environment: Content and features may be unstable or change without notice.
Advisory details: pypa / vllm/PYSEC-2026-3998
Advisory ID pypa/vllm/PYSEC-2026-3998
Aliases CVE-2026-93840
PYSEC-2026-3998
Summary 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.
Status Published
Exploitability 0.5
Weighted Severity 4.8
Risk 2.4
Affected and Fixed Packages Package Details
Introduced and Fixed Package Commit Patches Package Commit Patches Details
Weaknesses (0)
There are no known CWE.
No exploits are available.

No EPSS data available for this advisory.

Patch URL
There are no known patches.
Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:L/A:N Found at https://github.com/pypa/advisory-database/blob/main/vulns/vllm/PYSEC-2026-3998.yaml
Attack Vector (AV) Attack Complexity (AC) Privileges Required (PR) User Interaction (UI) Scope (S) Confidentiality Impact (C) Integrity Impact (I) Availability Impact (A)

network

adjacent_network

local

physical

low

high

none

low

high

none

required

unchanged

changed

high

low

none

high

low

none

high

low

none

There are no SSVC decisions available.