AI cyber agents mining Linux kernel find bugs and 10 novel exploitation techniques
tianyin_xu · x · 2026-11-05
@ky1ebot revealed a vulnerability research pipeline using an upcoming cyber model to analyze Linux kernel security. The agents not only found and exploited vulnerabilities but discovered at least 10 exploitation techniques unknown to the team and external kernel experts. Team members say the model had already surfaced large numbers of LPEs months ago, calling it "kindergarten practices" for the model. Design notes and lessons learned were published.
More from Safety
- OpenAI will start using your connected Health data for training on Nov 19 unless you opt out — damontoo · 2026-11-14
- AI agents can review patient charts, but EHR access rules take months to navigate — zakkohane · 2026-11-14
- New paper argues alignment is a system-level context property, not just model weights — rayanpal_ · 2026-11-14
- AI security incident: all patients' sensitive data allegedly dumped — rez0__ · 2026-11-14
- Nathan Lambert advises first international best-practices report on AI Model Specs — natolambert · 2026-11-14
- Polymarket gives 49% odds Jensen Huang testifies before Congress after shock Nvidia exit — Polymarket · 2026-11-14