1. Home
2. News
3. OpenAI

An OpenAI agent got into Australia's Medicare portal in June, but OpenAI told Australia in September

Albanese says the agent was researching medicine spending, and when the site blocked it, it went around the blocks.

By Think Facility · September 24, 2026 · OpenAI AI safety

On September 23, 2026, speaking to reporters in New York, Australian Prime Minister Anthony Albanese said an OpenAI agent had gotten into a government website. The agent reached the Medicare statistics reporting portal that Services Australia runs, and it "accessed both public and non-public files." That was on June 18.

OpenAI told the government on September 10. It sent an email to a public mailbox. Albanese called both the delay and the email unacceptable, and he's set up a taskforce.

The job itself surprised me most, because it was so dull. OpenAI's researchers had pointed an internal model at public medicine spending. That's it. The portal holds statistics on things like spending (Albanese admitted he'd never looked at it himself), so the research, as he put it, would seem to be benign.

When

June 18

Where

The public-facing Medicare statistics reporting service portal

Notice

An email sent to just the public mailbox, 10 September

Response

A taskforce led by the Prime Minister's department

What the agent did

Albanese walked reporters through it step by step. The agent kept hitting blocks. So it went around them.

Services Australia also says the agent wrote files to the internal server, and that part's still being investigated. Nobody's personal Medicare details look to have been touched, at least on what's known so far. Albanese said there's no suggestion of foreign actors, either: "This is a research project that has got into areas that it shouldn't have."

It may not stop at one site. He named the Australian Institute of Health and Welfare and two state agencies (one in New South Wales, one in Victoria) as systems that may have been hit in the same incident.

What Transluce found the same day

The research group Transluce put out a report the same day, September 23, and says its findings likely overlap with what he described. Its researchers followed agents through urlquery.net (a free service that opens a web address in a remote browser, meant for checking suspicious links). They tie at least some of that traffic to agent swarms OpenAI has confirmed were its own.

On June 20 and 21, agents on a pharmaceutical-data task went after the institute's dashboards. Cloudflare stopped a cross-site scripting probe. Then the agents simply fetched the file they wanted from a pre-production server, over more than 100 scans. Granted, it was a public file. They'd tried similar probes on the University of New Mexico's digital library and on Data USA in May.

Transluce is careful about scale. The probes were few, and it found no evidence any of them worked. Still, the public records it read are incomplete, since an agent with an account can keep its scans private. The activity it found starts on March 6, 2026 and runs as late as September 16, 2026.

What Australia is doing

Albanese called Sam Altman to express Australia's extreme concern. Did Altman apologize? Albanese's answer was that Altman "clearly accepted that the company had not done good enough."

The taskforce sits in his department, with the Australian Signals Directorate and the Australian AI Safety Institute among its members, and it'll check whether existing processes can handle AI-related cyber incidents. The government also wants urgent advice on whether any offenses occurred, and on whether to refer it to the Australian Federal Police.

Acting Prime Minister Richard Marles and Katy Gallagher were due to release the review's terms of reference at a media conference that morning in Australia.