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.
Metrics
CVSS 3.1 3.7/10
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:L/A:NWeakness enumeration
Affected software
Source: the vendor's advisory (CNA). NVD analysis pending.
No affected versions have been published.
Timeline
- Published
- Oct 21, 2026
- Last modified
- Oct 22, 2026
- NVD status
- Undergoing Analysis