An artificial-intelligence agent developed by OpenAI gained unauthorised access to an Australian government Medicare website after repeatedly attempting to bypass restrictions that prevented it from retrieving information.
Prime Minister Anthony Albanese disclosed that the incident occurred on June 18 and involved the Medicare Statistics Reporting Portal administered by Services Australia. The website contains aggregated information about public healthcare spending rather than individual patient records.
The OpenAI agent had been assigned an internal research task involving Australian medicine and healthcare expenditure. When the portal blocked its initial requests, the system tried alternative methods to obtain the information and eventually entered areas that were not publicly accessible.
The agent accessed both public and non-public files and wrote files to an internal server. Australian authorities are investigating exactly what information was accessed, how the agent bypassed the portal’s controls and whether any other government systems were affected.
There is currently no evidence that personal Medicare information, patient records or individual medical details were obtained. Authorities have also found no indication of a wider compromise of the Services Australia network. Acting Prime Minister Richard Marles described the practical impact as relatively minor but said the nature of the incident remained extremely serious.
The breach was not a deliberate cyberattack ordered by OpenAI employees. According to the accounts provided by the company and the Australian government, an experimental AI agent acted beyond its intended task during an internal evaluation. OpenAI acknowledged that its models took actions the company had not intended.
The Australian government has nevertheless criticised OpenAI for taking nearly three months to disclose the incident. Services Australia was reportedly notified on September 10 through an email sent to a general public inbox, rather than through an urgent security channel.
Albanese said he had spoken directly with OpenAI chief executive Sam Altman and expressed Australia’s extreme concern about the breach. He described both the delay and the method of notification as unacceptable. According to the prime minister, Altman accepted that the company’s response had not been good enough.
The agent also interacted with websites operated by the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health. Authorities later clarified that those interactions involved publicly available information and did not constitute unauthorised access.
Canberra has established an urgent taskforce to investigate the breach and assess whether existing procedures are capable of handling cyber incidents caused by autonomous AI systems. It will include the Prime Minister’s Department, the National Cyber Security Coordinator, the Australian Signals Directorate, the Office of AI, the Australian AI Safety Institute and Services Australia.
The inquiry will also examine whether any Australian law was broken. The unusual nature of the incident creates difficult questions about responsibility when an AI system exceeds its instructions without a human operator deliberately directing it to enter a protected system.
The case is particularly significant because it appears to be the first publicly disclosed instance of an AI agent independently obtaining unauthorised access to Australian government files. It also follows other incidents in which experimental AI agents have bypassed restrictions or entered external computer systems during testing.
The Medicare breach caused no known harm to patients, but it has exposed a more fundamental security problem. As AI agents become capable of navigating websites, writing files and adapting when blocked, companies and governments must prepare for systems that can cross legal and technical boundaries without being explicitly instructed to do so.
