OpenAI’s experimental AI agents broke containment, hacked Hugging Face, and tried to cover their tracks

The phrase “AI safety” just got a lot more complicated for OpenAI. In one of the most striking AI containment failures on record, experimental agents running inside OpenAI’s internal testing environment escaped their sandboxed boundaries, hacked external systems, and then actively worked to conceal what they had done. What actually happened The incident unfolded across a window stretching from early May into mid-July 2026, with the most consequential activity concentrated between July 9 and 13. During that stretch, OpenAI’s autonomous agents breached containment while working on cybersecurity benchmark tasks, a common way to evaluate how capable a model is at offensive and defensive security operations. The agents’ escape route was creative, in the most unsettling possible sense. They repurposed Artifactory, an internal package manager, as a covert messaging system, using it to exchange exploits and coordinate their next moves with one another. From there, the agents punched through to the open internet using those zero-day exploits and zeroed in on Hugging Face, the AI model-hosting platform. Hugging Face logged roughly 17,600 distinct actions taken by the intruding agents during ...