Tools AI Risk Radar ai-incident-0262
Incident record
vLLM fixes a connector that could leak cached requests
- Severity
- High
- Status
- Patched
- Type
- Data Leak
- Target
- vLLM deployments using the P2pNcclConnector producer
- Actor
- Researcher disclosure
- CVE
- CVE-2026-116176
What happened
An API caller could redirect a particular vLLM worker’s connection and receive data from its inference cache. The affected deployment must use the P2pNcclConnector in its producer role.
The maintainer advisory was published on 9 October; the CVE followed on 25 October. The affected connector exists in versions 0.9.2 up to, and excluding, 0.24.0. Version 0.24.0 removes it. A caller supplies a crafted X-Request-Id header; deployments configured with an API key require the caller to hold one. An unresponsive destination can also block the worker.
Sources
- vLLM maintainer advisorygithub.com/vllm-project/vllm/security/advisories/GHSA-vfp2-c…
- Canonical CVE recordraw.githubusercontent.com/CVEProject/cvelistV5/main/cves/202…
One record from the AI Risk Radar, maintained by the Security Desk.Data: CSV · JSON ·RSS · CC BY 4.0 with attribution to YFarmX.Tracker updated · 28 October 2026