A groundbreaking experiment by OpenAI has revealed the astonishing capabilities of artificial intelligence (AI) in discovering and exploiting Linux kernel vulnerabilities. The company’s unreleased cyber model, powered by an autonomous pipeline, was able to identify 948 unique Linux kernel flaws in a single run, which would be worth millions of dollars on the bug bounty market.
The mid-October scan targeted the latest code in the kernel’s master branch, where AI agents first narrowed down the kernel to security-relevant areas accessible by low-privileged users. They then flagged 2,308 potential bugs and confirmed 1,724 of them. The pipeline successfully tested 1,169 working local privilege escalation (LPE) exploits and 291 proofs of concept for information leaks on virtual machines running an unmodified desktop Linux setup with common protections like KASLR, SMEP, and SMAP enabled.
What’s striking about this experiment is the sheer number of vulnerabilities discovered. The 948 figure reflects the number of separate patches created for the bugs behind those exploits. Before submitting anything to kernel maintainers, OpenAI’s security researchers checked the exploits and fixes, along with the severity assessment suggested by the system. “To the best of our knowledge, the agents in this pipeline discovered and used at least 10 novel Linux exploitation techniques,” the company said.
Kernel exploitation experts outside OpenAI weren’t familiar with these new techniques, and for safety reasons, the full technical details are being kept private. The experiment’s cost was a staggering $1.2 million in API credits, which works out to roughly $1,279 per exploit. In comparison, bug bounty programs like Pwn2Own and TyphoonPWN have historically paid between $20,000 and $70,000 for Linux kernel privilege escalation exploits. This would put the 948 bugs at a value of $18.9 million to $66.3 million – 16 to 55 times what the scan cost.
Interestingly, OpenAI notes that this October scan uncovered more than five times as many new LPEs despite the kernel having undergone three months of hardening in between. The company also pointed its pipeline at an adapted version of Google’s kernelCTF and delivered 21 LPE exploits over 112 hours and about $33,000 in API credits.
OpenAI has already submitted hundreds of reports and patches for critical flaws and will send fixes for the remaining bugs in stages as part of its Patch the Planet initiative. The company also warned defenders that having access to a low-privileged account on an operating system without the latest available patches should be considered functionally equivalent to root access for a well-resourced threat actor.
To contain the damage when a flaw does get exploited, OpenAI suggests layered defenses like sandboxing and isolation technologies such as Firecracker and gVisor. Additionally, the company is increasing payouts in its Security Bug Bounty program to encourage outside researchers to hunt for flaws that its own testing may have missed. The goal is to get more eyes on potential vulnerabilities and ensure a safer online environment.
For users, this experiment serves as a stark reminder of the importance of staying up-to-date with the latest patches and keeping their systems secure. As AI becomes increasingly capable of discovering vulnerabilities, it’s crucial for individuals and organizations alike to prioritize cybersecurity and take proactive measures to protect themselves from potential threats.
Source: SecurityWeek — 2026-11-05