CVE-2026-114049 vLLM before 0.31.0 Stale allowed_token_ids Mask via InputBatch.condense

Medium 6.3 Published 21 Oct 2026 Updated 28 Oct 2026

Exploited in the wild: Not known
Fix: None listed

What is CVE-2026-114049?

CVE-2026-114049 is a medium-severity vulnerability in vllm-project vllm, published on 21 Oct 2026.

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.

Weakness: CWE-440

Which versions are affected by CVE-2026-114049?

How do I fix CVE-2026-114049?

No fix has been published for CVE-2026-114049 yet. Limit exposure of the affected software until one is available.

Is CVE-2026-114049 being exploited?

There is no known exploitation of CVE-2026-114049 in the wild and no public exploit code.

Sources

Also known as GHSA-6cxc-2vcg-w5qc, GHSA-qcgg-947w-p2vj.