CVE-2026-114049

Summary

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.

Impact & exploitability

CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:L/A:N

Affected products we track (1)

Recommendation

Apply the vendor fix in your normal patch cycle. Open any affected product above for its exact safe version.

Additional information

• NVD record
• https://github.com/vllm-project/vllm/security/advisories/GHSA-6cxc-2vcg-w5qcAdvisory
• https://www.vulncheck.com/advisories/vllm-before-0.31.0-stale-allowed-token-ids-mask-via-inputbatch-condenseAdvisory