OpenAI AI system finds nearly 1,000 Linux vulnerabilities and proposes fixes

OpenAI presented research showing how artificial intelligence could help protect systems used worldwide, from phones and online storage services to supercomputers. The study focused on the Linux kernel, the central component of the operating system that manages device operation and communication between programs and hardware.

In a single experiment carried out in mid-October 2026, the system analyzed 2,308 possible security issues in the Linux kernel. Of these, 1,724 were reproduced and confirmed under controlled testing conditions.

“Linux underlies much of the global digital infrastructure, including Android, online data centers, and supercomputers.”

The system demonstrated that 948 of the vulnerabilities could be used to obtain administrator rights, even by a user with initially limited access to the computer. It also identified 291 problems that could allow access to information that should have remained protected.

The experiment cost approximately 1.2 million dollars in credits for the use of OpenAI services. Linux underlies much of the global digital infrastructure, including Android, online data centers, and supercomputers.

The Linux kernel contains more than 40 million lines of code and is developed by an international community of programmers. Its size and complexity make it difficult to identify all security problems through traditional methods.

OpenAI's research followed the entire process, from identifying a possible vulnerability to proposing solutions for remediation. Automating stages of security research could help specialists analyze very large volumes of code and discover problems that would otherwise require lengthy investigations.