OpenAI model finds Linux kernel bugs and writes working exploits
An OpenAI autonomous cybersecurity model confirmed several hundred local privilege escalation flaws in the Linux kernel and produced 291 proof-of-concept exploits.
An autonomous cybersecurity model from OpenAI detected hidden bugs in the Linux kernel, verified them as real vulnerabilities, wrote working exploits and distributed patches automatically, according to digitaltoday.co.kr. The research was reported on November 5, 2026.
OpenAI researcher Eric Wallace said the single autonomous run changed the baseline for conventional network security. Out of 2308 initial detections, the model confirmed several hundred as local privilege escalation vulnerabilities, a category in which security experts typically find around 20 Linux flaws per year.
The model produced 291 working proof-of-concept exploits for memory leaks and documented at least 10 categories of previously unknown hacking techniques. API credit costs totaled $1.2 million, or $1,279 per completed exploit. Such exploits normally trade in the market for $20,000 to $70,000.
David Schwartz, honorary CTO of Ripple, reacted to the research results by saying he picked a good time to retire. In November 2025, Schwartz stepped down as Ripple CTO and moved to an honorary role, though he remains involved in development.
Quick answers
What did the OpenAI model find in the Linux kernel?
It flagged 2308 initial detections and confirmed several hundred as local privilege escalation vulnerabilities, then produced 291 working proof-of-concept exploits for memory leaks.
How much did the OpenAI run cost?
API credit costs totaled $1.2 million, or $1,279 per completed exploit.
Who commented on the research?
David Schwartz, honorary CTO of Ripple, said he picked a good time to retire.