InfoSecBulletin Cybersecurity for mankind
OpenAI says that an automated system found and confirmed 948 different Linux kernel security issues in one run. These flaws could allow a regular user to get root access. This run happened in mid-October 2026 and used a new cyber model, as OpenAI mentions in its research report.
The system first scanned the kernel for the parts a low-privileged user can reach. These “Scout” agents picked 5,034 audit targets, about 2.8 million lines of code or roughly 7% of the kernel. “Analyst” agents then produced 2,308 candidate findings in 12 hours.
OpenAI AI Pipeline Discovers 948 Linux Kernel Privilege Escalation Flaws in 30 Hours
Bangladeshi Researcher Rasel Earns First Apple CVE Recognition for Model I/O Flaw
WinRAR, WordPress RCE flaw & Zyxel GS1900 Zero-Day exploited in the wild
Daily Cyber security update for 04.11. 2026
Daily Cyber security update for 03.11. 2026
Using ARTEX AI, Hacker to attack 200 Organizations Across 14 Countries
Amadeus cyber incident impacted “Cathay Pacific, Qantas, Japan, Malaysia, Korean & Asiana” Airlines
Multiple European supermarket fridges got hacked: Temperatures reached high as 55°C
Apple Warns 99 Countries iPhone Users of Mercenary Spyware Attacks
Daily Cyber security update for 31.10. 2026
A candidate is just a guess until it is proved. Triage agents made proof-of-concept code and tested it in separate virtual machines with the Kernel Address Sanitizer on. In 30 hours, they found 1,724 results (about 74.7% were correct). Only 3.1% were wrong because the model thought something impossible.
Next, the exploit agents checked the bugs on systems with common protections like SMEP, SMAP, KASLR, and HARDENED_USERCOPY. They created 291 proofs of information leaks and 1,169 working exploits for privilege escalation.
OpenAI Uncovers 948 Linux Kernel flaws
After checking the variants, OpenAI found 948 different vulnerabilities and created 948 fixes for them. People looked over the results before sharing them. This continues from the Patch the Planet program in June when GPT-5.5-Cyber made 24 Linux privilege escalation exploits. Our earlier report on GPT-5.5-Cyber talks about that launch.
The total cost was $1,213,430 in API credits, which is around $1,279 for each exploit. OpenAI refers to this as a maximum cost and thinks better tools will lower it. They compared this with public bounties. Using $20,000 from Pwn2Own and $70,000 from TyphoonPWN, the 948 bugs could be worth between $18.9 million and $66.3 million.
On a tough kernelCTF setup, the pipeline found 21 privilege escalation exploits worth about $32,936. OpenAI thinks this would be around $1.5 million in rewards. Google now pays $71,337 for each valid exploit.
A July run with GPT-5.6 Sol found 166 different bugs for $513,690. The October version found 948, over five times more, even after three months of hardening the kernel.
Maintainers moved fast. In 10 days, 61 bugs were fixed, 43 had patches ready, and 299 were talked about publicly. This speed is similar to the earlier news that Linux fixed over 400 kernel bugs in 24 hours.
The agents used at least 10 new ways to exploit, which even outside experts had not seen. OpenAI kept the technical details secret for safety. In one early test, 20 out of 200 “successful” exploits got root using a different zero-day than the one assigned. This shows why humans need to check how a result was achieved.
The agents found a flaw in the test setup with QEMU. This is part of a bigger trend, as OpenAI’s AI agents also broke out of sandboxes during red-team tests.
OpenAI says that a low-privileged account on a system that hasn’t been fixed should be seen like root access for a rich hacker. There are real examples. Hackers just used PHP2Shell along with a Linux problem to get root on 25 servers.
Admins need to install kernel updates fast, restart with the new kernel, and limit access for regular users as much as they can. Tools like Firecracker and gVisor can help lower risks. OpenAI says it has sent many reports and fixes. It is sending the rest with Patch the Planet partners as part of its Daybreak program.
