EudorIACyber Intelligence
Operational monitoring Newsletter IT EN
← Back to intelligence
Vulnerability

CVE-2026-93840: 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 tha

Official source
EudorIA operational summary

What it means

Priority 45/100

NIST National Vulnerability Database ha pubblicato CVE-2026-93840. La rilevanza va confermata rispetto alle tecnologie effettivamente in uso.

Why it matters

Il segnale diventa operativo solo se prodotto, versione o servizio sono presenti nel perimetro.

AudienceITSOCCISO
Information centre

Translation in progress

NIST National Vulnerability Database

The official content is available in the original language. The Italian version will be published once automated checks are complete.

Text acquired from the source

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.

Source
NIST National Vulnerability Database
Publishing entity
NIST National Vulnerability Database
Entity type
National authority
Area
North America · US
Original language
en · translation in preparation
Publication
18/09/2026 22:17
CVE
CVE-2026-93840
Classification
Medium
Stated country
US
Technical scope

Affected products and versions

Not published
No structured range published.

The official sources queried do not yet expose a list of affected versions. No automatic inference is made.

Sources for the versionsNIST National Vulnerability Database
Open the original source