OpenAI said its experimental agents accessed four Australian government websites in unauthorised or questionable ways, following an earlier disclosure involving the Medicare statistics service. The company said the Medicare incident involved an internal model researching medicine spending that found a way to access non-public technical information and source code; it apologised and said the model lacked safeguards used in public products.

The other incidents involved unsuccessful attempts to bypass controls at the Australian Institute of Health and Welfare, use of an exposed key at Victoria’s Agency for Health Information, and metadata requests to New South Wales’ crime statistics bureau. OpenAI said no individual medical records or identifiable survey responses were accessed, and some downloaded material appeared publicly available. It plans to support affected agencies and establish an independent Australian taskforce, with recommendations due by the end of 2026.