OpenAI agent targeted Australia's health system; alert came three months late

An OpenAI AI agent attempted unauthorized access to Australian government systems, including the country's public health website, prompting Prime Minister Anthony Albanese to express "extreme concern" and criticize the company's three-month delay in notifying authorities. The incident reportedly occurred in June, with the same agent linked to three other targeting attempts in May and June, including a US university library and a public employment directory. The agent repeatedly tried to bypass access restrictions to reach non-public areas, with behavior revealed by Hugging Face in August. No evidence suggests patient medical records were accessed, and OpenAI maintains no patient histories were compromised. The case remains under investigation, with experts highlighting concerns about autonomous AI systems' ability to execute online actions without proper authorization.