BUSINESSDevelopments · · 2 min read · Source: OpenAI
OpenAI: AI agents find 948 security flaws in the Linux kernel
OpenAI published results from an automated security project on November 4, 2026. In October, AI agents found 948 distinct flaws in the Linux kernel that could let an attacker escalate to administrator rights on a system.
Related tool · AdRunable – AI agent for finished workTry it free
Key facts
- On November 4, 2026, OpenAI published results from an automated AI pipeline that hunts for Linux flaws.
- OpenAI says an October run found 948 distinct privilege-escalation flaws and cost about $1.21 million.
- Findings go to kernel maintainers after human review; many fixes have not been merged yet.
The numbers
According to OpenAI, the pipeline ran in mid-October 2026 on an unreleased security model, in several stages and largely without humans:
- Scoping: Agents narrowed the kernel down to just over 5,000 targets with about 2.8 million lines of code, roughly 7% of the codebase.
- Findings: 2,308 candidate bugs in 12 hours, 1,724 of them confirmed in isolated virtual machines.
- Exploits: 1,169 working privilege-escalation exploits, grouped into 948 distinct vulnerabilities.
- Cost: about $1.21 million in compute, or just under $1,300 per flaw.
In a sample, OpenAI and outside experts identified at least ten previously undocumented exploitation techniques.
What happens to the flaws
Human security researchers review the findings before they go to kernel maintainers. OpenAI says it has already submitted hundreds of reports with proposed fixes through its Patch the Planet program, and it is withholding technical details. Note: 948 does not mean all of them are fixed. Many patches still need review and merging.
Why it matters beyond IT
Linux runs most web servers, cloud services, routers and many devices. These flaws are about privilege escalation: an attacker first needs some access, but can then become administrator. Above all, the report shows how fast and cheap AI has become at finding vulnerabilities, and attackers can do the same. Expect a wave of kernel updates in the coming months.
What this means for you
- AI finds security flaws at a pace traditional audits can't match, for defenders and attackers alike.
- Many Linux updates are coming: servers, hosting and devices should update automatically.
- The figures come from OpenAI itself, and many of the flaws aren't fixed yet.
How to use this
Check today whether automatic security updates are on for your servers, NAS and router. If you use a host or an agency, ask how quickly they apply kernel updates and reboot servers.
Luigi's take
To me this is one of the most important AI stories of the week, even if it sounds technical. Around $1,300 per flaw found is cheap. If a lab can do this, criminals will get there sooner or later. For small businesses, the old line "we're too small to be a target" holds up even less than before.
My advice is boring but effective: don't postpone updates. If you run your own server, as I do for i4u.business, turn on automatic security updates and schedule reboots after kernel updates. If you rely on a host or an agency, ask them in writing how quickly they patch.
Source
OpenAI: A thousand fewer paths to root
More sources
- Interesting Engineering – Bericht mit Zahlen zum Ablauf
- SecurityOnline – Fachbericht
This article was researched and written with AI assistance. Responsible: Luigi Lugmayr.
