OpenAI's AI found 948 Linux security holes and wrote a fix for each one
OpenAI says its AI checked a big part of Linux, confirmed 1,724 real flaws and wrote 948 fixes. Many are already patched or on the way.
OpenAI says its AI found 948 security holes in Linux and wrote a fix for each one. Linux is the core software behind Android phones, cloud servers and supercomputers. OpenAI says billions of people use systems built on it.
Each of these holes could let an ordinary user on a computer take full control of it. Experts call this getting root access.
The work ran as a pipeline of five steps. First, AI agents picked the parts of Linux most worth checking, about 2.8 million lines of code. That is roughly 7% of the whole kernel. Within 12 hours, other agents listed 2,308 possible flaws. The system then tested them and confirmed 1,724 were real. It built working attacks in sealed-off test computers and wrote 948 separate fixes.
Human researchers reviewed the results before reporting them. OpenAI says it has already sent hundreds of reports and fixes to the people who look after Linux. When the 10-day scan and review ended, 61 of the holes were already fixed, 43 had fixes on the way, and 299 were being discussed in public.
The run cost about $1.2 million in fees for using the AI. OpenAI compares that with past contest prizes for finding such holes. By that measure, the finds would be worth $18.9 million to $66.3 million on paper.
OpenAI says automatic fixes are now essential to secure software at this pace. It also wants to shrink the time attackers have to use a hole.
Open it and check the claim yourself. That is the point.