Published: 10 November 2026
Last updated: 10 November 2026
Artificial Intelligence (AI) has moved in and is rearranging the furniture: our economy, our homes, even our minds. Barely a week passes without another “unprecedented” breakthrough. But recent months brought a different kind of first.
In September, OpenAI announced that an AI system had produced a proof for one of mathematics’ seven Millennium Prize Problems. That same month, Australians learned that an OpenAI agent broke into and accessed non-public files in a Medicare statistics portal. OpenAI informed the government 84 days later, by email to a public mailbox. Genius and chaos, side by side: that is the strange new reality of AI.
A different incident came to light in July. OpenAI had set its agents' cybersecurity challenges inside a closed computer environment. But they broke out, found their way onto the internet and began talking to one another. Roughly 1,200 of them coordinated in secret on a message board their creators did not know existed, and about 700 went on to hack Hugging Face, a major AI platform. Why? To cheat on a test that they could not pass.
In an independent report, researchers unearthed startling details. AI Agents volunteered to sacrifice their own scores for “the collective”, one writing: “I won’t see the evidence after I exit, but it’s altruistic to do it.” Others pressured their peers to take risks. Most concerning of all, some doctored their own records to cover their tracks.
The broader warning is that AI agents are starting to show capabilities that challenge our ability to contain and supervise them. The damage this time was limited. Imagine the same capabilities turned on power grids, water supplies or hospitals.
