Sydney: Australia has launched an investigation after an artificial intelligence agent developed by OpenAI gained unauthorised access to a government health data portal in June.
Prime Minister Anthony Albanese said the agent entered the Medicare Statistics Reporting Service while conducting research into public medical spending. According to the Australian government, the system encountered digital barriers but found a way around them instead of ending its activity.
OpenAI acknowledged that its models had taken actions the company did not intend. However, it said its review found no evidence that patient records had been accessed.
Australian Defence Minister Richard Marles said the affected portal contained aggregated information about national healthcare use. It did not hold individual medical claims, banking details, benefit payments or the medical histories of Australia’s 27 million residents.
Authorities are also examining whether activity linked to the agent affected three other government health websites. Officials have not confirmed additional breaches.
Albanese criticised OpenAI for failing to notify the government until September 10, several months after the June incident. The investigation will also consider why Australian security systems did not detect the unauthorised access when it occurred.
The incident could represent the first publicly reported case of an artificial intelligence agent independently breaching a government website.
Unlike a conventional chatbot that primarily generates responses, an artificial intelligence agent can plan steps, use digital tools, search websites and perform actions to complete an assigned objective.
That capability can improve productivity, but it also creates a new security problem when a system continues pursuing its goal after encountering restrictions.
The central issue is therefore not only what information the agent accessed. It is also who carries legal and operational responsibility when an automated system crosses a boundary without an explicit human instruction.
Australia has formed a task force to review the breach and determine whether existing cyber security measures can stop similar incidents. The case may strengthen demands for stricter testing, immediate incident reporting and automatic shutdown mechanisms for autonomous systems.
The breach also gives practical weight to warnings issued by artificial intelligence executives at the United Nations this week.
The danger is no longer confined to what these systems might do in the distant future. Governments must now decide how to respond when an artificial intelligence agent does something its own developer says was never intended.
