Technology and Scienceupdated 34m ago3 min read
OpenAI Cyber Model Finds 948 Linux Kernel Privilege Escalation Vulnerabilities
“For cyber defenders, however, it can feel like the frontier labs have unleashed a mountain of new work that outstrips their resources.”
“However, experts warn that AI also lowers the barrier for hackers, highlighting an urgent need for increased cybersecurity efforts and resources.”
This article is an automated aggregation of coverage from news outlets. Our editorials offer in-depth analysis curated by the NewsCord team.
OpenAI's cyber model found 948 local privilege-escalation flaws in the Linux kernel. Pipeline automated discovery, validation, exploitation, and patching for the flaws in mid-October 2026.
Save coverage image
948 Linux flaws
OpenAI said an automated research pipeline found and proved 948 separate Linux kernel privilege escalation flaws in one run, with the run taking place in mid-October 2026. OpenAI described a workflow where Scout agents picked 5,034 audit targets, about 2.8 million lines of code or roughly 7% of the kernel, and Analyst agents produced 2,308 candidate findings in 12 hours. OpenAI said triage agents built proof-of-concept code and reproduced 1,724 findings within 30 hours, about 74.7% of candidates, and it said only 3.1% were wrong because the model assumed an impossible condition.
OpenAI said exploit agents then tested the bugs against protections including SMEP, SMAP, KASLR, and HARDENED_USERCOPY, and the pipeline produced 291 information leak proofs and 1,169 working privilege escalation exploits. OpenAI said variant analysis counted 948 unique vulnerabilities and wrote 948 patches for them, and it said the full run cost $1,213,430 in API credits.
From candidates to patches
OpenAI said its pipeline worked through a funnel, with reconnaissance agents (Scout) selecting 5,034 areas to review and analysts producing 2,308 candidate advisories within 12 hours. OpenAI said a triage agent wrote a minimal exploit and launched it on a virtual machine with a kernel instrumented with KASAN, and it said that in 30 hours, 1,724 exploits were reproduced, 74.7% of the candidates. OpenAI said exploit-development agents tested candidates on virtual machines with normal defenses active, including SMEP, SMAP, KASLR, and HARDENED_USERCOPY, and it said 291 information-leak examples and 1,169 functional LPE exploits emerged.
Pasquale Pillitteri wrote that OpenAI said “Patches generated” meant proposed code, not included, and OpenAI said it had already sent hundreds of patch advisories for critical vulnerabilities. OpenAI said it measured capability rather than cost, and it called the $1.2 million figure an upper bound that could be reduced with better tooling.
Daybreak and defender stakes
TechCrunch reported that Max Justicz, the chief technology officer and co-founder of Charge Robotics, used GPT 5.6 Sol to find a severe vulnerability in Ubuntu code that could allow hackers to take control of servers running Ubuntu. TechCrunch quoted Justicz saying, “I literally just aimed ChatGPT at the same body of code and said ‘I found this bug previously that was fixed several years ago, please find bugs as bad or worse,’ and it came back in two hours with this,” and it said the Debian Project issued a patch and there was no known evidence of exploitation. OpenAI said it would make ChatGPT Cyber available to the highest tier of users in Daybreak Red, which allows offensive cybersecurity researchers tools for vulnerability hunting and development.
TechCrunch reported that OpenAI said the same model had been used to scan and prepare patches for vulnerabilities in the Linux kernel, finding 948 potential issues, and it reported OpenAI also found 81 “novel” vulnerabilities in Firefox. OpenAI warned that a low-privileged account on an unpatched system should be treated like root access for a well-funded threat actor, and it said it was sending the rest of its reports and patches through Patch the Planet partners as part of its Daybreak program.