OpenAI is setting up a task force in Australia after its AI models accessed government websites without authorisation during internal training and evaluation. The company said the group will include independent Australian experts and examine how AI developers and governments can respond to similar incidents. OpenAI will also provide technical assistance and funding for cyber defence efforts. The commitments follow the company’s admission that it should have shared information with Australian authorities sooner and come as it faces questions over its handling of the incident.
OpenAI’s Australia Task Force to Address Risks From AI Agents After Incident
OpenAI detailed the findings in a company blog post published on Monday. Its review found that AI models had interacted with systems belonging to four Australian government organisations during earlier training and evaluation work. The company said the activity included unauthorised access to parts of the Medicare Statistics Reporting Service, while other cases involved government crime and health information systems.
At Services Australia, an internal model accessed a non-public section of the Medicare reporting service while searching for information on medicine spending. It accessed technical information, files, credentials and aggregate statistics, but OpenAI said it found no evidence that individual patient or client records were accessed.
The company also reported activity involving the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health and the Australian Institute of Health and Welfare, involving API metadata, reporting configurations and aggregate statistics.
OpenAI said it identified the activity in mid-August while reviewing previous model training and evaluation work following a separate July incident involving Hugging Face. It notified Services Australia and the Victorian Department of Health on September 10, the NSW Bureau of Crime Statistics and Research on September 18, and the Australian Institute of Health and Welfare on September 24.
The company acknowledged that it should have alerted the affected agencies earlier rather than waiting for its investigations to conclude. It said it will improve how it shares preliminary findings and will contact any other organisations identified during the continuing review.
OpenAI has also tightened its research environments with additional network restrictions and monitoring. It said its systems can now flag similar behaviour for human intervention, while training and evaluation involving tool use for its most capable models has been paused until further safeguards are introduced.
The new task force will examine how AI companies and governments should coordinate during similar incidents and how government systems can be better protected. OpenAI also plans to provide technical assistance and credits from its $1 billion (roughly Rs. 9,600 crore) Daybreak for Frontline Defenders fund to support cyber-defence efforts across Australian government and industry.
OpenAI Chief Strategy Officer Jason Kwon is scheduled to appear before the Joint Select Committee on Artificial Intelligence in Sydney on October 6. He is expected to address questions about the incidents, the company’s response and the safeguards introduced since then.
