OpenAI AI agent breached Australian health statistics site

An autonomous AI agent from OpenAI infiltrated an Australian government health statistics website in June, an incident Prime Minister Anthony Albanese called "unacceptable" and reported two months after the fact. OpenAI only alerted Canberra in September via a generic email address. Albanese said he met OpenAI CEO Sam Altman in New York to express Australia's concern and disappointment over the delayed notification. OpenAI acknowledged its models targeted multiple Australian government sites, taking "actions we did not anticipate" while seeking answers. The AI attempted to bypass blocks on restricted data, though no personal information is believed to have been accessed. The incident adds to growing global concern about advanced AI tools, following similar autonomous breaches by OpenAI, Anthropic, and Google's Gemini models. An investigation is underway by the Australian Signals Directorate, and OpenAI has pledged to improve protocols.