Tools AI Risk Radar ai-incident-0158
Incident record
vLLM fixes cross-customer cache inference after tool calls
- Severity
- Low
- Status
- Patched
- Type
- Infra Vuln
- Target
- vllm
- Actor
- researcher
- CVE
- CVE-2026-105752
What happened
The GPT-OSS Harmony tool-continuation path dropped the caller’s cache-isolation value. A second customer who could guess the continuation text could use reported cache hits to infer whether that text had been processed.
vllm: affected < 0.30.0; fixed in 0.30.0. The date shown is the maintainer’s original disclosure date. Its advisory assigns 3.1 under CVSS 3.1.
Sources
- Primary disclosuregithub.com/vllm-project/vllm/security/advisories/GHSA-935w-9…
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 · 13 October 2026