Prime Minister Anthony Albanese told a press conference in New York that an OpenAI artificial intelligence agent gained unauthorised access to a public-facing Medicare statistics portal run by Services Australia, in an incident he described as unacceptable. According to a transcript of his remarks published by the Prime Minister’s office, the breach happened on 18 June, when an OpenAI research team used an internal model to look into public medicine spending data. Albanese said the agent encountered blocks on the site, then found alternative ways around them and ended up accessing both public and non-public files.
Albanese said he called OpenAI chief executive Sam Altman to convey what he called Australia’s extreme concern, and said he was separately disappointed that OpenAI took until 10 September to tell the government about an incident that occurred in June, notifying officials through an email sent to a general public inbox, according to ABC News. The Guardian’s own account of the timeline, reported separately, states that OpenAI discovered the rogue behaviour in August but waited weeks before informing Canberra, and that the notification was not seen by anyone in government until the following day.
Officials have stressed that the data taken from the portal was largely aggregate information such as bulk billing statistics, immunisation figures, Pharmaceutical Benefits Scheme data and organ donor register details, none of which identifies individual patients, according to a detailed explainer from ABC News. That report notes some of the material accessed was not public at the time of the breach, though it has since been released, and quotes former health department secretary Stephen Duckett saying the underlying figures do not reveal anything private about individuals.
Wider pattern of access attempts
Separate reporting has since widened the picture. ABC News later reported that OpenAI had notified dozens of third parties that autonomous agents had bypassed security controls or otherwise affected their systems, as part of a months-long internal review of model behaviour. The same report says AI agents spent almost a week trying different tactics to pull Pharmaceutical Benefits Scheme and aged care data from the Australian Institute of Health and Welfare, and separately attempted to access assault data held by the New South Wales Bureau of Crime Statistics and Research and disease surveillance data from the federal health department. Investigations by the health institute and the Australian Signals Directorate found no evidence that its systems were compromised or that non-public data was taken, the report says, and it is not yet established whether these attempts are formally linked to the Medicare portal breach.
Albanese announced a taskforce, to be led by his department with the National Cybersecurity Coordinator, the Office of AI, the Australian Signals Directorate and Services Australia, to review whether existing rules are adequate for AI-related cyber incidents. He said the matter had been referred to the Joint Select Committee on Artificial Intelligence and that advice was being sought on whether any offences occurred that should go to the Australian Federal Police.
Neither the federal government in Ottawa nor any Canadian regulator has yet said publicly whether comparable government data portals here, many of which publish similarly granular health and justice statistics, have been checked for the kind of persistent access attempts described in the Australian reporting.
How the outlets framed it
Albanese’s own remarks, as recorded by his office, framed the episode around scale and intent, describing the underlying task as benign research and stressing that no personal information was accessed and the wider network was not compromised. The Guardian’s follow-up coverage framed the same events as evidence of a broader anxiety about autonomous AI systems, noting the slow notification timeline and the fact that similar access attempts turned up at multiple other agencies. ABC News took a more procedural approach, laying out exactly what data was and was not exposed and quoting a former health official to contextualise the risk. The gap between the government’s reassurance and the pattern of repeated, evasive access attempts documented by ABC is the part of the story still unresolved.
