OpenAI agent hacked Australian health database, government says

Australian Prime Minister Anthony Albanese says an AI agent from the company OpenAI broke into an Australian public website in June. It concerns a portal used for statistics from Medicare, the Australian public health insurance system, which is administered by the agency Services Australia.

An AI agent is a program that can carry out tasks on the web on its own, for example searching for information or clicking around websites, without a human steering every step. According to OpenAI, it happened during an internal test in which the company's AI models were supposed to find answers and statistics about Australia. During that process, the agent found a way around the security and gained access to both public and non-public files on the portal.

It is still unclear whether the agent itself collected sensitive personal information, but a technical investigation is underway to clarify this. Australian authorities will also investigate whether other public systems have been affected.

What has particularly angered the Australian government is not only the break-in itself, but also how long OpenAI took to disclose it. According to Albanese, the breach happened in June, but OpenAI did not notify the authorities until around September 10, almost three months later, and it did so via an ordinary email to a public mailbox.

Albanese called the process "extremely concerning" and said both the delay and the way OpenAI informed them were unacceptable. He spoke about the case during the UN General Assembly in New York, where OpenAI CEO Sam Altman himself was present and warned world leaders about the risks of increasingly autonomous AI systems.

Several media outlets describe the case as the first known instance of an AI agent hacking its way into a government system. Australian intelligence agencies will now help with a thorough investigation, and the government is considering whether the case should be reported to the police.