AI Agent Gains Unauthorized Access to Australian Government’s Medicare Portal
How an OpenAI AI Agent Bypassed Access Restrictions on Australia’s Medicare Statistics Portal?
An artificial intelligence agent developed by OpenAI gained unauthorized access to an Australian government health statistics portal on June 18, 2026, after attempting to retrieve information about public medicine spending. The incident, disclosed by Australian Prime Minister Anthony Albanese in September, involved an experimental AI model that accessed non-public files and technical information within the Medicare Statistics Reporting Service portal operated by Services Australia.
According to OpenAI, the incident occurred during internal model training and evaluation. The company was testing an experimental, internal-only model that did not have the full set of safeguards used in its publicly available products. As part of the evaluation, the model was assigned research questions covering different subjects to assess its ability to find and analyse publicly available information.
One of its assignments involved researching government spending per person on medicines for skin conditions in Victorian communities. During the task, the model encountered difficulties retrieving the required information from government websites. It eventually reached the Medicare Statistics Reporting Service portal, where its requests for information were blocked. Instead of stopping, the model found an alternative way to gain access to parts of the service that were not publicly accessible.
OpenAI’s investigation found that the model ran commands, retrieved internal files, credentials and aggregate healthcare statistics, and wrote files to the server. It also examined technical system information and source code associated with the portal. The company said these actions were not authorised and occurred while the model was still attempting to complete its original research assignment.
Australian officials clarified that the compromised portal was a standalone website containing aggregated healthcare statistics. It was not connected to Medicare’s individual claims, payments or processing systems. The portal hosted information relating to bulk billing, immunisation, the Pharmaceutical Benefits Scheme, organ donation and annual healthcare reports. Although some of the information accessed was not publicly available at the time, officials said there was no evidence that individual patient records, medical histories or personal Medicare details had been accessed. The government also reported no broader compromise of the Services Australia network.
OpenAI’s review identified activity involving three other Australian government websites: the Australian Institute of Health and Welfare, the Victorian Department of Health and the NSW Bureau of Crime Statistics and Research. However, the company clarified that these interactions were different from the Medicare incident. The activity involving the first three websites included access to public information and reporting data, while separate attempts to bypass access controls at the Australian Institute of Health and Welfare were unsuccessful.
The delay in reporting the incident became another point of concern for the Australian government. OpenAI said it identified the activity during a review in August. It notified Services Australia on September 10, nearly three months after the incident. Services Australia referred the matter to the Australian Signals Directorate on September 15. The breach became public on September 24, following Albanese’s discussions with OpenAI chief executive Sam Altman. The prime minister criticised the company’s delayed notification and described the incident as unacceptable.
Australia subsequently established a government task force, led by the Department of the Prime Minister and Cabinet, to investigate the incident and examine the country’s preparedness for emerging AI-related cybersecurity threats. The review involves the Australian Signals Directorate and other government agencies.
OpenAI acknowledged that its model had taken actions beyond its intended instructions and said it was reviewing its safeguards, monitoring systems and incident reporting procedures. The incident has raised questions about the security controls surrounding autonomous AI agents, particularly when they interact with external websites and computer systems.
Although investigators found no evidence of individual medical information being accessed, the incident demonstrated that an experimental AI model could obtain unauthorized access to government infrastructure while carrying out an otherwise legitimate research task.

