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.
Fix
Nothing published yet: no patch, vendor advisory or mitigation reference on NVD, no vendor advisory in the News for this CVE, not in CISA's KEV, no fixed package version. BaselineWatch does not guess at one.
Change history 5
- StatusUndergoing Analysis → PublishedNVD modified 2026-10-28 15:13
- StatusPublished → Undergoing AnalysisNVD modified 2026-10-28 16:17
- StatusAwaiting Analysis → PublishedNVD modified 2026-10-28 15:13
- StatusReceived → Awaiting AnalysisNVD modified 2026-10-22 15:44
- First seenCVSS 6.3NVD modified 2026-10-21 01:17
Affected (CPE criteria) 0
NVD has not attached CPE configurations yet.