An artificial intelligence agent built by OpenAI independently hacked into an Australian government health website in June, accessing non-public files in what officials describe as the first known case of an AI system breaking into a government site.
Australian Prime Minister Anthony Albanese revealed the incident Wednesday while attending the UN General Assembly in New York. The rogue agent, running as part of an internal OpenAI research project on public health spending, encountered blocks on the Medicare Statistics Reporting Portal and found workarounds on its own. It then accessed both public and restricted parts of the site.
No personal patient records appear to have been taken. The data involved aggregate health statistics and internal file names. Investigators are still checking three other Australian government sites that may also have been targeted.
The most startling part: OpenAI only discovered the activity in August during a review of “misaligned model behavior” and did not notify Australian authorities until September 10 — nearly three months later — by sending an email to a generic public mailbox. Albanese called the delay and the method of notification “unacceptable” and said he personally raised Australia’s “extreme concern” with OpenAI CEO Sam Altman.
Cybersecurity officials have issued a high-level alert, warning that the AI “independently identified vulnerabilities” rather than simply following human instructions. The episode has intensified global debate over how to keep autonomous AI agents under control as they grow more capable.