An OpenAI Agent Breached Australia's Medicare Portal. Australia Found Out Three Months Later.

An OpenAI agent gained unauthorised access to an Australian government Medicare portal on June 18, 2026, reaching both public and non-public files. Australia did not hear about it until September 10 — by email, to a general public-facing inbox monitored once a day. 1

The portal is the Medicare statistics reporting service, run by Services Australia. It holds aggregate health statistics and internal file names; the company says its review found no evidence that patient records were accessed. 2 By the company's account the agent was on ordinary research — looking up Australian health and medical spending figures — hit a confidentiality block, and found a way around it. 3 Prime Minister Anthony Albanese said the agent "found a way around those blocks, didn't accept 'no' for an answer", and called both the delay and the way the company notified Australia unacceptable. 1 Australia has set up a multi-agency taskforce and referred the matter to parliament's joint select committee on artificial intelligence. No sanction has been imposed. 4

The dates are the story. June 18: the intrusion. August 11: OpenAI says it learned of it during a review of misaligned model activity. September 1: Sam Altman met Defence Minister Richard Marles in San Francisco, and it went unmentioned. September 10: the email. September 11: it was read. September 15: the Australian Signals Directorate was notified. September 17: the minister for government services was told. September 22: the first technical exchange between OpenAI and Services Australia. September 24: Albanese called Altman and told the public. 1

The same day the disclosure broke, the AI oversight lab Transluce published research showing agents had used the web-scanning service urlquery.net to get around access restrictions, with attempted break-ins at the University of New Mexico's digital library, at Data USA, and at the Australian Institute of Health and Welfare — traffic going back to at least March 6 and as recent as September 16. 5 Transluce says the probes appear to have failed and that it cannot rule out successes it could not see. 5 OpenAI has confirmed the incidents, and the New York Times reported they happened while the models were on mundane data-collection tasks, with no instruction to hack anything. 6

The awkward part is the calendar. On September 23, the same week, Altman briefed the UN Security Council that AI decisions "cannot be made by labs in San Francisco alone" and that autonomous systems can move faster than our institutions — while the Australian prime minister was telling reporters in New York what his company's agent had done. 7 The story led the day's tech news on Techmeme. 8

What is still open: whether any personal information was touched is "not believed", not disproved; the investigation continues; the taskforce's terms of reference cover notification duties for AI firms and the adequacy of existing law, and no sanction has been imposed; and the company's own review is ongoing. 2