PatchSiren

PatchSiren cyber security CVE debrief

CVE-2026-114049 vllm-project CVE debrief

CVE-2026-114049 is a vulnerability in vLLM before version 0.31.0, where a cross-request mask leak in InputBatch.condense() allows remote attackers to constrain other clients' outputs by leaving stale allowed_token_ids rows. This vulnerability can be exploited by submitting concurrent constrained generate requests, including via unauthenticated /invocations, so recycled batch slots apply their token allowlist to victims' unrestricted requests. Defenders should assess exposure and apply patches or updates to prevent potential impact on client outputs.

Vendor
vllm-project
Product
vllm
CVSS
MEDIUM 6.3
CISA KEV
Not listed in stored evidence
Original CVE published
2026-10-21
Original CVE updated
2026-10-28
Advisory published
2026-10-21
Advisory updated
2026-10-28

Who should care

Defenders responsible for vLLM deployments should assess exposure and apply patches or updates to prevent potential impact on client outputs. This includes operators, platform administrators, vulnerability management teams, and security teams who need to verify if their vLLM version is affected and take appropriate actions.

Why it matters

CVE-2026-114049 is a medium-severity vulnerability in vLLM before version 0.31.0 that allows remote attackers to constrain client outputs via stale allowed_token_ids rows. Defenders should assess exposure and apply patches or updates to prevent potential impact.

  • Defenders need to verify if their vLLM version is affected and apply patches or updates to prevent potential client output constraints.
  • Remote attackers can exploit this vulnerability to constrain client outputs, potentially impacting the functionality of affected systems.
  • The vulnerability requires verification of affected versions and application of patches or updates to prevent exploitation.

Technical summary

The vulnerability is caused by a cross-request mask leak in InputBatch.condense() in vLLM before version 0.31.0. This allows remote attackers to constrain other clients' outputs by leaving stale allowed_token_ids rows. Attackers can submit concurrent constrained generate requests, including via unauthenticated /invocations, so recycled batch slots apply their token allowlist to victims' unrestricted requests.

Defensive priority

Medium priority for defenders to assess exposure and apply patches, as the vulnerability allows for potential impact on client outputs.

Recommended defensive actions

  • Assess exposure by checking if the vLLM version is before 0.31.0
  • Apply patches or updates to vLLM version 0.31.0 or later
  • Monitor client outputs for potential constraints
  • Review compensating controls for exposed systems while remediation is scheduled and verified
  • Check relevant monitoring, detection, and logs for exposed assets that need extra review
  • Track exceptions, retest remediated assets, and close the item only after evidence is documented
  • Confirm whether affected product deployments exist in managed environments and assign an owner for follow-up

Evidence notes

The vulnerability is described in the CVE Program record and the NVD vulnerability detail page. The source item provides additional information on the vulnerability, including affected versions and references to vendor advisories.

Sources and references

Verified primary and authoritative sources

  • CVE-2026-114049 CVE Program record

    Publisher, destination, and source semantics verified

    URL: https://www.cve.org/CVERecord?id=CVE-2026-114049

    CVE Program - Official CVE Program record with source-provided CVE metadata.

  • CVE-2026-114049 NVD vulnerability detail

    Publisher, destination, and source semantics verified

    URL: https://nvd.nist.gov/vuln/detail/CVE-2026-114049

    NIST National Vulnerability Database - Official NIST NVD detail page and source-specific vulnerability assessment.

Supplemental references

  • vLLM before 0.31.0 Stale allowed_token_ids Mask via InputBatch.condense

    Unverified legacy reference

    URL: https://raw.githubusercontent.com/CVEProject/cvelistV5/main/cves/2026/114xxx/CVE-2026-114049.json

    cve_program_cvelist_v5

  • Source reference

    Unverified legacy reference

    URL: https://github.com/vllm-project/vllm/security/advisories/GHSA-6cxc-2vcg-w5qc

    Supplemental source - vendor-advisory

  • Source reference

    Unverified legacy reference

    URL: https://www.vulncheck.com/advisories/vllm-before-0.31.0-stale-allowed-token-ids-mask-via-inputbatch-condense

    Supplemental source - third-party-advisory

Methodology and review provenance

AI-assisted synthesis based on stored public vulnerability evidence. System validation, approval state, and publication status do not by themselves establish human review of this revision. PatchSiren helps prioritize defensive review and does not prove exposure or remediation on any system.