948 Exploits Against the Linux Kernel: When AI Turns Bugs into Attacks—and Writes the Fixes

Last updated:
Written by:
Techguidebe
TechGuide.be Verified Publisher

Table of Contents

 




An OpenAI model that has not yet been released exploited the Linux kernel 948 times in different ways.


In a research post, OpenAI revealed that an automated system running on an upcoming AI model of its own turned hundreds of bugs in the Linux kernel into actual attacks that give an ordinary user full root privileges on the machine, then wrote a fix for each vulnerability.


The experiment took place in mid-October 2026 on the latest version of the kernel. It cost $1,213,430 in usage credits, or about $1,279 per attack. According to the researchers, the automated agents used 10 exploitation techniques that were not previously known even to independent experts.


The company chose Linux because it runs Android devices, the cloud, and the supercomputers that billions of people rely on. By the end of the testing, 61 vulnerabilities had been officially fixed, while OpenAI continues to send reports and fixes to the kernel developers.


The most important message, according to the researchers: an unupdated machine should be considered fully exposed to a resourceful attacker, so updating your systems and applications remains the first line of defense.


Do you think AI’s ability to discover vulnerabilities protects us more, or opens a more dangerous door?

Author & Blogger

Author and blogger on this site, dedicated to providing exclusive and useful content.

Comments