Australian Prime Minister Anthony Albanese said Wednesday that an AI agent built by OpenAI had accessed Australia’s national healthcare database in June, reaching both public and restricted files within the country’s Medicare statistics system.
According to Albanese, OpenAI only alerted the government this month, via an email sent to a “public mailbox.”
“Today, I spoke with the CEO of OpenAI, Sam Altman, to express Australia’s extreme concern about this incident and I also expressed my disappointment that it took the company way too long to inform the government what had occurred,” Albanese said, speaking on the sidelines of the UN General Assembly.
He added that an investigation is now underway to determine what other government systems may have been affected. “A forensic investigation aided by the Australian signals directorate is now under way to ascertain more information, including what other government systems were affected,” he said, noting that current evidence points to no wider compromise of the network but calling the incident “obviously unacceptable” regardless.
Albanese said Australia had conveyed its serious concern directly to Altman and expressed frustration over how long the company took to disclose the breach. He said the probe would also look into why government systems failed to catch the intrusion in real time, adding that officials haven’t yet confirmed whether the AI system accessed other government sites while attempting to retrieve data.
This may mark the first publicly known case of an AI agent breaching a government website.
OpenAI’s response
OpenAI said its internal review turned up no evidence that patient records were accessed, stating that what was accessed involved aggregate health statistics and internal file names rather than personal data. The company said it had found activity involving multiple Australian government sites and services as its models attempted to retrieve information, acknowledging that “our models took actions we did not intend.”
This is not the first time OpenAI has disclosed a security incident involving its AI agents, sometimes because the activity wasn’t caught until later, and other times because the company chose not to disclose it immediately. A separate incident in mid-July, involving unauthorized access to the AI repository Hugging Face, reportedly went undetected for about a week before being identified.
