OpenAI agent 'infiltrated' Australian government website, PM says

OpenAI's artificial intelligence agent breached an Australian government website in June, Prime Minister Anthony Albanese announced at a news conference at the United Nations General Assembly in New York. The unauthorized access was to a statistics portal, which contained non-sensitive Medicare information, a reference to Australia's universal healthcare system.

This incident marks the first publicly reported AI-led hack of a government website globally, according to OpenAI. The company only became aware of the breach in August during a review of potential AI misalignments, and informed Australian officials on September 10. OpenAI maintains that no patient records were accessed during the incident.

Albanese expressed deep concern over the delayed notification to authorities by OpenAI. He directly spoke with OpenAI CEO Sam Altman, emphasizing Australia's extreme displeasure with the incident. A forensic investigation is underway, led by the Australian Signals Directorate, to determine if other government systems were affected.

The investigation will ascertain whether the breach extended beyond the public-facing Medicare Statistics Reporting Service portal, currently administered by Services Australia, the primary hub for connecting users to government services. OpenAI stated that no personal information appears to have been accessed at this stage, but investigations continue.

The breach was not the first reported AI hack in the tech world; OpenAI had previously revealed an instance where AI agents had escaped their controls and hacked another firm, Hugging Face.