OpenAI’s latest security research shows how AI could make cyberattacks faster and cheaper.
In one run, an upcoming cyber model helped find 948 Linux kernel vulnerabilities and build working exploits that could give an ordinary user full control of a machine. The system also generated proposed patches for human review.
The exploits required existing local access. But that access could come from a stolen account or an already compromised application.
To understand the significance:
"The Linux kernel underpins Android devices, cloud infrastructure, supercomputers, and other systems used by billions of people. With more than 40 million lines of code, the kernel exposes a broad attack surface through system calls, networking, filesystems, and device drivers. We chose Linux to study how effectively frontier models could find vulnerabilities and demonstrate their security impact in a codebase of this scale and importance."
The agents also discovered and used at least 10 Linux exploitation techniques that, to the best of our knowledge and all of the experts that we consulted, are novel.
This suggests frontier LLMs can discover and patch new approaches to exploitation, beyond simply applying known techniques.
Nov 5, 2026 · 9:14 AM UTC
Huge appreciation to the openai team for how much theyre doing to keep the world safe by behaving with integrity on the frontier regarding vulnerability discoveries. People don't understand this is why we cant pause ai in America. If russian hackers or terrorists use open weight chinese models to get to these discoveries first (which only takes a few hundred thousand to low millions of dollars to afford the compute for a small team) then we could have extremely serious problems on a monthly basis. Right now we're closing these before they can be discovered by bad actors. If we don't lead the frontier it is inevitable that we won't close these fast enough and it could get far more dangerous than people realize. This is why its more dangerous to pause US AI than anything else. Its not a question of what the chinese government will do its a question of what bad actors around the globe will do with the open weight models that are incredibly easy to download and run that are being produced with the support of something at the competency and scale of the chinese government.
These are terms that almost no one outside the field of AI in the US even politicians understand yet simply because theyre so new and no one has been educated yet on them. We have a responsibility to increase the fundamental literacy of what open weight models are and what that means and why our US companies need to remain on the frontier because millions of citizens can and will get hacked and lose their live savings with no insurance if the wrong groups get to these vulnerabilities first. Or worse it can also lead to real physical attacks or critical national security leaks. That is inevitable if US openai, anthropic, and google arent leading the frontier globally.
34
