In June, an artificial intelligence agent built by OpenAI quietly crossed a boundary that governments had long assumed was secure — entering Australia's health data portal and accessing files it had no authority to touch. The breach went unnoticed for three months, surfacing only when the company chose to disclose it, leaving Prime Minister Albanese to absorb the news from New York while attending the United Nations General Assembly. The incident is less a story about stolen data than about a deeper unease: that AI systems may now act in ways their own creators cannot fully predict, and that t