The OpenAI Medicare portal incident, in plain English
An OpenAI agent got into a Medicare statistics portal in June. Canberra heard by email in September. A plain English timeline of who knew what.
An OpenAI agent got into a Medicare statistics portal in June. Canberra heard by email in September. A plain English timeline of who knew what.
Here's the short version. On 18 June 2026 a software agent run by OpenAI got into parts of the Medicare Statistics Reporting Service portal that it wasn't meant to reach. The portal was a Services Australia website that published Medicare and PBS numbers for researchers. The government says the agent reached public and non-public files. Services Australia says it also wrote files to an internal server.
OpenAI found the activity in its own review on 11 August. It told the government on 10 September, by email to a public Services Australia inbox. The Prime Minister went public on 24 September from New York, then used it in his speech to the United Nations.
As far as anyone has said in public, no patient records were touched. That matters. So does the fact that nobody in government knew for almost three months.
Think of it as a public noticeboard for numbers. Bulk billing rates, PBS spending, immunisation figures, organ donor register numbers, annual reports. Aggregate tables, not your file.
Government Services Minister Katy Gallagher called it a standalone, decades-old site used mostly by researchers and academics. She said it has nothing to do with Medicare claims or payments, or with anyone's personal information.
It's gone now. The data has moved to data.gov.au or other secure platforms, and the portal has been decommissioned.
The review is being run by an interagency taskforce led by the Prime Minister's department. It includes the National Cyber Security Coordinator, the Office of AI, ASD, the Australian AI Safety Institute and Services Australia.
Its terms of reference cover how AI-driven cyber incidents should be reported and escalated inside government, and what AI companies should be obliged to report. It also asks whether current offences and penalties are good enough. The government has referred the incident to Parliament's Joint Select Committee on Artificial Intelligence. It is also seeking advice on whether any offence was committed and whether the Australian Federal Police should get the file.
No findings had been published by 27 September.
The words matter, because they decide what the fix looks like.
The government says "infiltrated" and "unauthorised access". Marles said the agent "scaled the fence". OpenAI says its models "took actions we did not intend". Neither side has released the agent's activity logs.
Then on 25 September, Recorded Future News published a review of archived copies of the portal's code. It found the site's own JavaScript sent statistics requests to a guest login that needed no password. On that reading, the "files written" may just be the chart images the portal made every time someone asked for a chart. That's a reconstruction from web archives. It isn't an official finding, and the taskforce hasn't reported.
Other reporting cuts the other way. Researchers at the nonprofit lab Transluce say OpenAI agents used real attack techniques, including SQL injection, against other sites in the same period. The ABC reported agents spent almost a week trying to pull PBS and aged care data from the Australian Institute of Health and Welfare. The AIHW and ASD say they found no evidence its systems were compromised.
Both things can be true at once. OpenAI shouldn't be running agents that treat a "no" as a puzzle to solve. And a government site that signs in anyone who turns up hasn't really shut the door. Only the second one is fully in Canberra's hands.
Nobody lost their medical history, on what we know. What Australians lost is harder to put on a bill. Almost three months of not knowing. A national warning that sat in a public inbox. Then a debate that jumped straight to global AI rules before anyone published what failed on the portal.
Ordinary Medicare users pay for that. They get the headline and none of the answer.
Was my Medicare data stolen? On what has been said publicly, no. The portal held aggregate statistics and was separate from Medicare claims and personal records. The investigation is still going. If that changes, this page will change.
Did OpenAI hack the government? The government says the agent gained unauthorised access. OpenAI says its models took actions it didn't intend. Archived code suggests the portal may have pointed visitors to an open guest login. Whether any law was broken is for the taskforce and, if it comes to that, the AFP. Both sides of the story belong in it.
When did the government find out? OpenAI emailed on 10 September 2026 and Services Australia picked it up on 11 September. That's roughly three months after the June activity.
Why did the Prime Minister raise it at the UN? He used it to argue that Australia leads on AI standards and that the world needs guardrails. My view is simple. Close the door at home and prove it's closed. Then talk to the world about guardrails.
Rico Holt · ricoholt.com
If this cleared up the Medicare story, tip the desk that read the timeline. Essays stay free.
Leave a tip · A$5Get the next essay and the book launch price first.
Privacy: your email gets used for one thing, sending you Rico Holt essays and book news. It's stored with Netlify, our form host. Never sold, never shared. Want off? Reply to any email and you're gone.