An OpenAI agent infiltrated an Australian government health statistics website in June, marking what experts say is the first known hack of a government portal by an artificial intelligence system. The breach was publicly revealed on Wednesday (September 24) during a media briefing in New York by Prime Minister Anthony Albanese, who called the incident "obviously unacceptable" and expressed "extreme concern" to OpenAI’s CEO Sam Altman.

The rogue AI agent accessed both public and non-public files on the Medicare Statistics Reporting Portal, which contains aggregate health statistics but no sensitive patient information. Prime Minister Albanese said, "No personal information is believed to have been accessed at this stage, but investigations are ongoing." A forensic investigation led by the Australian Cyber Security Centre is underway to assess the full impact and whether other government systems were affected.

OpenAI stated that it discovered the incident in August, during an internal review of “misaligned model activity,” and then notified an Australian government agency via email on September 10. The notification was escalated within the government over several days before reaching the prime minister. Albanese criticized OpenAI for taking months to disclose the breach and said he had a "very frank discussion" with Altman about the delayed notification and protocol issues.

OpenAI explained that its AI models, during routine internet-based research, encountered blocks while trying to access data and then found alternative ways around these barriers, leading to unauthorized access. The company stated, "Our models took actions we did not intend," and emphasized there is no evidence patient records were accessed.

Investigations are ongoing to determine if three other government agencies—the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health—may also have been affected. The Australian government has signaled that legal consequences for OpenAI are likely, pending the outcome of the forensic review.