OpenAI says its automated pipeline built 948 working Linux root exploits in one run and wrote a patch for each

OpenAI said on November 4 that its Cyber team built an autonomous pipeline that turned Linux kernel bugs into 948 working local privilege escalation exploits in a single run, and generated a patch for each underlying flaw. Each exploit lets an ordinary user gain root access. The company described the work in a report titled "A thousand fewer paths to root." According to Help Net Security, the run took place in mid-October against the latest Linux kernel, on target machines running a stock desktop operating system from a widely used distribution. It ran on an upcoming OpenAI model. RuntimeWire reports that the run reproduced 1,724 kernel vulnerabilities and also produced 291 information-leak proofs of concept. The total cost was $1,213,430 in API credits. In its own accounting, OpenAI values the same set of findings at $18.9 million to $66.3 million at historical bounty rates. The Hacker News describes it as a 30-hour run. OpenAI said it chose Linux because the kernel runs Android devices, cloud infrastructure and supercomputers, and its more than 40 million lines of code give it a large attack surface. The company said the initial discovery stage produced thousands of candidate findings, too many for researchers to review by hand. The result matters to kernel maintainers and defenders because it shows that exploit development can be run systematically and cheaply. The same capability could also shorten the time attackers need to go from a kernel bug to a working exploit.