A Human Found the Linux Bug AI Missed

AI is getting surprisingly good at finding security flaws—but it still isn't catching everything.

Earlier this year, Anthropic's AI model discovered a real vulnerability in the Linux kernel, and developers quickly patched it. The problem seemed solved.

Then security researcher Jaeyoung Chung examined the exact same code and found another flaw hiding nearby.

Known as Bad Epoll, the bug allows an ordinary user to gain full root privileges through an extremely subtle race condition. Even more impressive, Chung built an exploit that reliably triggered the vulnerability despite the attack window lasting only a few CPU instructions. The flaw can affect both Linux systems and many Android devices.

The good news is that Linux developers fixed the issue months ago, and there is no evidence it has been exploited in the wild. The bad news is that many devices still haven't received the update.

The bigger takeaway goes beyond this single bug. AI is becoming a powerful tool for finding vulnerabilities, but this case is a reminder that one successful AI audit doesn't guarantee software is secure. Sometimes the most dangerous flaw is the one the machine never noticed.