OpenAI's cyber model found ~1000 Linux privilege escalation flaws in a week for $1.2M

JiaweiLiu_ · x · 2026-11-05

OpenAI announced an autonomous cyber vulnerability discovery, exploitation, and patching pipeline, allocating significant frontier compute to pro bono patching of open source codebases.

In one run, an upcoming cyber model validated and patched 948 Linux kernel local privilege escalation (LPE) vulnerabilities for $1.2M in API costs
Before AI, researchers typically found 20 LPEs per year in Linux; the model found 1000 in a week and at least 10 novel exploitation technique classes
OpenAI says Linux is load-bearing software for the company, hence the priority

Original post →

OpenAI's cyber model found ~1000 Linux privilege escalation flaws in a week for $1.2M

More from Models