OpenAI Found Thousands of Vulnerabilities in Linux Using AI

OpenAI spent $1,213,430 on a security research project for the Linux kernel using artificial intelligence.

AI agents analyzed the source code and identified 2,308 potentially vulnerable areas. Of these, 1,724 were confirmed as real vulnerabilities, and working exploits were created for 948 of them.

Automated systems not only found vulnerabilities but also generated patches to fix them, sending the corrections to the kernel developers.

The average cost per exploit was $1,279. Experts estimate that similar work at competitions like Pwn2Own would have cost between $18 million and $66 million.

Experts warn that in the future, token limits may become the only barrier against automated AI hackers.