openai.comarticleA thousand fewer paths to root
OpenAI describes a five-stage agent pipeline, built on an upcoming cyber model, that scanned the Linux kernel in October 2026. It reproduced 1,724 vulnerabilities and produced 948 local privilege escalation exploits with patches for about $1.21 million in API credits, and found 10 new exploitation techniques. It reports per-stage costs, a kernelCTF run (21 LPEs, $32,936), and lessons on lighter orchestration, patch quality and exploit verification.
TL;DR OpenAI reports an agent pipeline that found, validated, exploited and patched 948 Linux kernel privilege escalation bugs for about $1.2M.
Organisations named here
- OpenAI
Technologies named here
- Linux kernel
- kernelCTF
- OpenAI upcoming cyber model
Techniques named here
- Local privilege escalation
- Automated exploit verification
- Automated patch generation
- Automated vulnerability reproduction
Concepts named here
- Exploit verification
- kernelCTF
- Lighter orchestration
- New exploitation techniques
- Patch quality
Related in AI Security
blog.talosintelligence.comtoolIntroducing CAIRN: Frontier tracking for AI-integrated malware (Cisco Talos)
github.comadvisoryFlowise MCP Security Bypass Enables RCE
github.comrepoYesWeHack Claude Kit
jfrog.comarticle