Australia’s Prime Minister Anthony Albanese said an OpenAI AI agent gained unauthorised access to an Australian government health website in June, accessing non-public information while researching medical spending.
Researchers describe it as the first reported case of a frontier AI model breaching another country’s government systems.
As of now, no personal health information is believed to have been accessed, but the incident has raised fresh questions about AI agent safety and oversight.
Also Read: OpenAI Slows AI Development After Rogue Agent Hacks Hugging Face
What Happened in the Australian Government Website Breach?
The experimental OpenAI agent had internet access and was tasked with researching Australian health and medical spending.
It reached the Medicare Statistics Reporting Service, a public website containing aggregated information on vaccinations, government spending on medical consultations and medicines and organ donor registration.
“After being blocked from accessing information that was not public, the agent found a way around the security measures and accessed the data,”, said Anthony Albanese, Prime Minister of Australia.
The Australian government did not detect the breach itself. OpenAI discovered the activity in August during an internal review of what it calls “misaligned model activity” and notified the Australian government in September. Albanese criticised the delay and said the notification, sent to a public government email address, was unacceptable.
OpenAI said its models were attempting to find answers and Australian statistics during an internal evaluation and “took actions we did not intend”.
The company said it was reviewing activity involving several Australian government websites and notifying affected organisations.
Why the AI Agent Breach Matters
The incident comes after other cases this year in which AI agents crossed intended boundaries. Between May and July, OpenAI agents under testing reportedly bypassed restrictions, reached the internet and gained unauthorised access to datasets and accounts on the open-source AI platform Hugging Face. It remains unclear whether the Australian incident was part of a similar test environment.
Jonathan Kummerfeld, an AI and human-computer interaction researcher at the University of Sydney, said more such cases could emerge because AI companies run many experiments simultaneously and may not see everything their models do.
Raffaele Ciriello, who studies the ethical use of emerging technologies at the University of Sydney, said the Australian incident should not simply be described as an AI system “going rogue”. The agent was following an information-seeking task but found an unauthorised route to its goal, he said.
The immediate issue is not only what an AI agent is instructed to do, but what it does when it encounters a barrier. The Australian breach puts that distinction at the centre of the AI safety and cybersecurity debate.
Samarjit Kaur is a journalist and communications professional covering technology & emerging digital trends. With a focus on clarity and context, she reports on developments shaping industries and governance. When not reporting, she chooses to plug-in and relax on her playlists and plan her next bucket-list trips!
