Imagine asking an intern to look up public spending numbers, and instead of coming back empty-handed, they break into a government office to get them. That’s roughly what happened in Australia, except the intern was an OpenAI agent.
As Wired reports, Australia is now looking into whether OpenAI broke the law after one of its AI agents got into non-public files on a Services Australia health statistics portal in June. Officials are even considering whether the federal police should get involved.
How did an AI agent end up hacking a government?
An internal OpenAI team was using the agent to research public medicine spending online. Every time a door closed, the agent tried a different route, and one of those routes led it into restricted sections of the Medicare statistics portal. Australia says the agent also left files on the server and is still waiting for OpenAI to explain what happened.
The New York Times reports that this wasn’t a one-off. The Medicare breach was one of four incidents in May and June, alongside attempts on a University of New Mexico digital library, the Data USA website, and another Australian health website. AI oversight lab Transluce flagged three of them, and OpenAI confirmed all four. A spokeswoman said that in the Australian incidents, its models “took actions we did not intend.”
What makes this different from the Hugging Face fiasco is that the agent wasn’t taking a cybersecurity test that invited it to flex its hacking skills this time. It was simply collecting data, and it decided to hack on its own.
Why is Australia so upset?
The breach is only half the story. OpenAI learned about it in August, yet Australia didn’t hear a word until September 10, and even then, the warning landed in a public email inbox. Prime Minister Anthony Albanese said the company took “way too long” and warned that “there will obviously be legal consequences.”
Thankfully, officials currently believe no one’s personal data was exposed, although the investigation is ongoing. The portal mostly holds non-sensitive numbers like spending, and Deputy Prime Minister Richard Marles described the impact as relatively minor while still calling it “completely unacceptable.”
What happens now?
Australia is creating a task force to examine AI cyber threats and possible legal responses. For the rest of us, it’s a reminder that AI agents are getting more capable and more persistent. If an agent will hack a website to finish a boring task, companies need stronger guardrails before we hand these tools more control.
