OpenAI spent $1.2m on Linux kernel vulnerability search

AI agents found nearly a thousand ways to hack the Linux kernel

OpenAI announced the results of its system code analysis. Artificial intelligence found 948 working exploits for the Linux kernel. The total cost of this development was $1,213,430.

During the scanning, the algorithms analysed the kernel code and identified 2,308 potentially dangerous sections. Detailed analysis showed that 1,724 of these are real vulnerabilities. For 948 cases, the AI agents managed to generate functional exploits.

Fixes handed over to developers

In addition to finding weaknesses, the artificial intelligence prepared a corresponding patch for each detected vulnerability. These fixes have already been sent to the Linux kernel development team.

The average cost of developing one ready-made exploit under this methodology is $1,279.

Economic efficiency of AI in cybersecurity

To gauge the scale: at professional cybersecurity competitions, such as Pwn2Own, tens of thousands of dollars are paid for similar findings. If OpenAI valued these exploits at general market prices, the cost of the work could have reached between $18m and $66m.

It seems that soon the only protection against AI hackers will be the token limit.