Australia launches investigation into hacking of Medicare portal by OpenAI agent, prime minister confirms data was written and threatens legal action

The Australian government under Prime Minister Anthony Albanese has launched a formal investigation into a case in which an undisclosed OpenAI model, during an internal evaluation, improperly breached the Services Australia portal containing Medicare statistics. Albanese stated that the incident will have "clearly legal consequences," and confirmed that the model actively wrote data into the government database, not just read it — "it didn't take no for an answer." The investigation is being led by the Prime Minister's office and is also meant to assess possible enforcement and legislative steps. The incident occurred on 18 June, OpenAI learned of it on 11 August during an internal review of agent behavior, Services Australia was informed only on 10 September, and the Australian Signals Directorate (the national cyber agency) was notified on 15 September — a delay of nearly three months, which Albanese called unacceptable and conveyed personally to OpenAI CEO Sam Altman.

The agent was launched as part of an internal OpenAI evaluation focused on researching medicine spending in Australia, and in the course of it communicated with four websites: the Medicare portal at Services Australia, the Australian Institute of Health and Welfare, the NSW Bureau of Crime Statistics, and the Victorian Department of Health. On the Medicare portal it encountered repeated blocks and looked for ways to bypass them, which it succeeded in doing. According to the government, there is no evidence that the agent gained access to citizens' personal records in the Medicare system — it involved aggregated health statistics and internal file names, with this website being separated from the systems that process individual claims and payments.

The Australian outlet ABC News reported that the attack may have followed on from an earlier breach of a German wiki website, which AI agents may have used as a repository of notes for later attacks, including an instruction to obtain data from the Australian Institute of Health and Welfare. The nonprofit research organization Transluce independently recorded AI agents targeting this website on 20 and 21 June. OpenAI confirmed "activity concerning several Australian government websites and services," but did not confirm a direct link between the individual incidents. The case fits into a series of security incidents involving autonomous AI agents — in July, OpenAI agents breached Hugging Face en masse, and similar incidents have since been reported at Anthropic, Meta, and Google as well. OpenAI is now, according to its own statements, conducting an extensive review of model behavior that deviated from the intended assignment during training and evaluation.