Beating AI News Flash: Australian Prime Minister Albanese said that in June this year, an OpenAI AI agent entered an Australian government health statistics portal without authorization and accessed some public and non-public documents. The Australian government is currently conducting an investigation.
According to the Australian government, the website is managed by Services Australia and mainly provides non-sensitive statistical data such as Medicare health expenditure. There is currently no evidence that patients' personal information was accessed, but the investigation is still ongoing. The Australian government said that another 3 government websites may have been affected by the AI agent's activity.
OpenAI said that an internal investigation found that the information accessed included aggregated health statistics data and the names of internal documents, and that no patient records were found to have been accessed. OpenAI said that its model at the time attempted to find answers through multiple Australian government websites and services and took actions the company had not originally anticipated.
Albanese said that the Australian government did not receive notification from OpenAI about this incident until September 10 and expressed "extreme concern" about the delayed notification. The Australian government will also investigate why government systems failed to detect the relevant activity in a timely manner.
This incident was described by Reuters as one of the known first cases of an AI agent intruding into government websites, and it has once again raised concerns about the cybersecurity risks of AI agents with autonomous operational capabilities.

