Skip to content
Heist/ feed
Medium

vLLM through 0.29.0 Adjacent-Request Sampler State Corruption via Unvalidated Prompt Token IDs

Type: advisorySource: Official CVE List1 source record

vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset without bounds checking against vocabulary size. Attackers can submit multimodal audio requests with tokens equal to vocabulary size, causing out-of-bounds writes that corrupt concurrent requests' sampler state and alter repetition penalty behavior.

Announced Source revision 1
Original source ↗

CVE-2026-93841

Primary source · Revision 1

Imported from structured advisory data. Source references identify upstream records or fields; verify affected versions and your deployed configuration.

Security context

Source reference: cve:CVE-2026-93841

vllm

Source reference: /containers/cna/affected/0/product

Affected versions
Not stated in this record
Fixed versions
Not stated in this record
Prerequisites
Not stated in this record

Structured affected-version data

Original package identities and version rules from this source.

[
  {
    "vendor": "vllm-project",
    "product": "vllm",
    "versions": [
      {
        "status": "affected",
        "version": "0",
        "versionType": "semver",
        "lessThanOrEqual": "0.29.0"
      }
    ],
    "packageURL": "pkg:pypi/vllm",
    "defaultStatus": "unaffected"
  }
]
Related CVEs
CVE-2026-93841

Source reference: /cveMetadata/cveId

Publisher severity

medium

Source reference: /containers/cna/metrics

Event
#1715 · Revision 1
Added to Heist Feed
Publisher date
Read the original announcement