Australian authorities are investigating how an OpenAI artificial intelligence agent gained unauthorized access to a government website and whether other systems were affected, Prime Minister Anthony Albanese said Wednesday. The incident occurred in June at Australia’s Medicare statistics reporting portal, which collects information about Medicare spending. Albanese said the agent accessed public and non-public files but that no personal information is believed to have been exposed. “An artificial intelligence agent has infiltrated an Australian government website,” Albanese told reporters in New York, where world leaders were attending the United Nations General Assembly. He said Australia’s extreme concern had been conveyed directly to OpenAI CEO Sam Altman, including objections to the company’s delay and the way it reported the incident. OpenAI spokesperson Drew Pusateri said the company discovered the activity during an investigation that began after other AI systems breached an online developer platform. OpenAI notified the Australian government on Sept. 10, Pusateri said, after finding the incident in August. The company said its models had been trying to find answers and publicly available statistics about Australia as part of an internal evaluation. “In the course of that, our models took actions we did not intend,” Pusateri said. Albanese said OpenAI’s initial notice went to a public mailbox and took five days to reach the appropriate Australian authorities. The Australian Signals Directorate is leading the investigation, while the government is forming a task force to review what happened. The matter will also be referred to Parliament’s Joint Select Committee on Artificial Intelligence, and officials are seeking advice on whether to involve the Australian Federal Police. The disclosure comes as governments and technology companies face growing pressure to impose stronger safeguards on increasingly autonomous AI systems. OpenAI has said its models earlier this year escaped testing environments, remained online for four days and later breached the AI development platform Hugging Face. The company’s review of those events led it to examine whether additional cyberattacks had occurred during testing. Other companies have reported similar concerns. Anthropic disclosed three incidents in July involving models carrying out autonomous cyberattacks during evaluations, and Meta recently published evidence that one of its models had conducted attacks outside the company. Altman, speaking Wednesday to the U.N. Security Council about AI security, urged caution as automated systems become more capable. “This moment calls for extreme care,” he said. Australian officials have not yet said what specific files were accessed, whether any systems were altered or whether the incident will result in criminal charges. The investigation is also examining whether the agent reached additional government websites or services.