Vulnerability in vllm-project vllm

[ Description ]Official description

vLLM before 0.31.0 contains a cross-request mask leak in InputBatch.condense() that 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.

[ Fixes ]Fixes and patches

No fix or interim patch is linked from our sources yet. Check the vendor references below and limit exposure of affected systems until a fix is available.

[ Advisories ]Government CERT advisories

• BSI CERT-Bund20 Oct 2026WID-SEC-2026-4076

  vllm: Multiple vulnerabilities Translated from German

  CVE-2026-114050CVE-2026-114053CVE-2026-114052+4 more CVEs

[ References ]References

• github.com/vllm-project/vllm/security/advisories/GHSA-6cxc-2vcg-w5qc
• www.vulncheck.com/advisories/vllm-before-0.31.0-stale-allowed-token-ids-mask-via-inputb…