Posting on its blog on Nov. 4, OpenAI explained how it decided to focus on finding potential paths of attack in the Linux kernel. This powers Android devices and cloud infrastructure used by billions of people, and with over 40 million lines of code, it offers an incredibly broad attack surface.
As part of that, OpenAI developed “an autonomous vulnerability discovery, exploitation, and patching pipeline. In one run with our cyber model, it validated and generated patches for 948 local privilege escalation [LPE] vulnerabilities in Linux at an API cost of $1.2 million.” OpenAI said this included at least 10 exploitation techniques that, to its knowledge, had not been seen before.
This vulnerability detection pipeline was set to work on the Linux kernel inside a protected environment. There, the AI analyzed 2,308 potential findings, reproduced 1,724 vulnerabilities, and managed to create almost a thousand local privilege escalation exploits that, if properly harnessed, could grant an unprivileged attacker root access to the system.
The pipeline is “powered by an upcoming cyber model, covering vulnerability discovery, dynamic validation, exploitability assessment, and patch generation,” OpenAI said. Human operators then analyzed the discoveries and reported the flaws alongside any relevant patches.
Given the rate at which frontier models are discovering vulnerabilities, OpenAI said it believed that “automated patch generation is essential to secure software at the pace required, and to limit the exploitation window.”
Linux maintainers are already moving quickly. When the 10-day scan and analysis finished, 61 of the flaws had been patched upstream, 43 had patches in staging and 299 were being reported and discussed publicly. OpenAI said it has sent hundreds of reports and patches for critical vulnerabilities.
Raising the costs to attackers
It’s not entirely clear which Linux distribution was tested or what model OpenAI used to find the exploits, but the company noted that in June it had used GPT-5.5 Cyber to identify Linux kernel vulnerabilities and had created 24 related exploits to validate the findings.
The API cost to OpenAI of finding these vulnerabilities shows how cheap this kind of research has become — and the huge savings that make identifying these exploits worthwhile for defenders. At a total cost of $1.2 million and 948 LPE flaws found, the cost per LPE exploit was just $1,279.
For context, the Pwn2Own and TyphoonPWN hacking contests have historically offered between $20,000 and $70,000 per Linux kernel exploit. That works out to between $18.9 million and $66.3 million, or 16 to 55 times OpenAI’s pipeline costs.
OpenAI put the risks in stark terms, saying that “having access to a low-privileged account on an operating system that isn’t up-to-date on its latest available patches should be considered functionally equivalent to root access for a well-resourced threat actor.”
