By Renju Jose and Chris Thomas
SYDNEY, Sept 23 (Reuters) - Australia said on Wednesday an OpenAI agent breached a government health data portal in June, gaining unauthorised access to files, in what could be the first known instance of an AI agent hacking a government website.
The breach is one of the highest-profile incidents of AIagents accessing external systems outside the United States,coming on top of several recent breaches globally by rogue AIagents that have alarmed governments and companies.
Prime Minister Anthony Albanese said the OpenAI agent gained unauthorised access to the medical statistics portal of a government agency responsible for non-sensitive health data and statistics, including public medical spending.
"Evidence currently available is there is no broader compromise to the ... network. Nonetheless, this situation isobviously unacceptable," Albanese said during a media briefing in New York, where he is attending the UN General Assembly. Investigations continue and Australia had voicedits "extreme concern about this incident" to OpenAI CEO SamAltman, Albanese said, adding that he was deeply disappointed by the company's delay in notifying the government. "It took until September 10 before there was anynotification at all," Albanese said, adding that theinvestigation would also examine why government systems hadfailed to detect the breach in the first place.
He also warned that three other government websites "may beimpacted" by the OpenAI agent's activity.
"The question is, when it was trying to harvest data, did itgo into these other sites? So we're not confirming that thatoccurred," he said.
The incident comes after OpenAI and Anthropic, in separate submissions to a parliamentary inquiry this month, urged Australia to reconsider a ban on use of the country's creative content to train their models.
'AI MODELS ATTEMPTED TO LOOK UP ANSWERS' "Our review found no evidence of patient records being accessed. The information accessed included aggregate health statistics and internal file names," OpenAI said in a statement.
It added that it "identified activity involving several Australian government websites and services as our models attempted to look up answers ... our models took actions we did not intend."
The Australian breach is the latest of several recent incidents in which OpenAI has disclosed hacks or unauthorised activity involving its AI agents well after they occurred.
In some cases this was because the activity was only detected belatedly, and in others because the company initially elected not to disclose it. A separate high-profile incident, a mid-July intrusion into open-source AI repository Hugging Face, was only detected about a week after it took place, according to timelines released by OpenAI and independent investigators.
This helped ignite global debate about the risks posed by increasingly powerful AI models.
Rivals Anthropic, Google's Gemini, and Meta have also disclosed incidents of their agents accessing external systems.
Some of America's top AI executives, including Altman, have called for a slowdown of industry development, citing, among other factors, the threat of devastating cyber attacks by out-of-control agents.
(Reporting by Renju Jose in Sydney, Chris Thomas in Mexico City, Gnaneshwar Rajan in Bengaluru, Raphael Satter in Washington; Editing by Maju Samuel, Jonathan Spicer and Clarence Fernandez)
