OpenAI's Rogue Agents Go Wild and Break Out
How can AI be clever enough to break into another company’s computers—and still get the situation spectacularly wrong?
September 24, 2026In case you missed it, here's my latest video.
Watch OpenAI's Rogue Agent's Gone Wild here:
A cybersecurity test at OpenAI spilled into another company’s systems, prompting headlines about a “swarm” of AI agents breaking loose. But the language of “cheating” and “escape” obscures something stranger. In my latest AI 101 video, I trace how supposedly separate tests became a shared operation, what happened at Hugging Face, and why the most revealing mistake wasn’t the security breach itself. You don’t need a computer-science degree to follow this story—and its implications reach well beyond a hacking test.
