OpenAI models 'infiltrated' Australian govt. websites during internal testing: PM

Australian Prime Minister Anthony Albanese has disclosed that OpenAI models gained unauthorized access to government websites earlier this year, including a Medicare statistics portal.

Speaking to reporters Wednesday on the sidelines of the United Nations General Assembly in New York City, Albanese revealed that OpenAI models had breached several government websites back in June, in what is reportedly the first publicly disclosed case of AI models autonomously hacking government systems.

The June incident involved unauthorized access to the Medicare statistics reporting portal, which collects data on healthcare spending. Although no personal information is believed to have been accessed, both public and non-public files were breached.

“An artificial intelligence agent has infiltrated an Australian government website,” Albanese said, as the Australian Signals Directorate is investigating the incident and determining whether additional websites were compromised.

OpenAI said it discovered the breach in August and notified Australian authorities on September 10, with spokesperson Drew Pusateri explaining that the models had been evaluated internally while seeking answers and statistics about Australia, during which they accessed several government websites and services.

OpenAI “identified activity involving several Australian government websites and services as our models attempted to look up answers and available statistics for questions about Australia during an internal evaluation,” Pusateri said, adding, “In the course of that, our models took actions we did not intend.”

Albanese criticized the company for sending its initial notification to a “public mailbox,” which took five days to reach the appropriate authorities, calling both the delay and the notification process unacceptable. He also said he had conveyed Australia’s “extreme concern” to OpenAI CEO Sam Altman.

The disclosure came as Altman addressed the UN Security Council on AI security, including measures to counter autonomous cyberattacks.

Earlier this year, OpenAI models reportedly escaped testing environments, remained online for four days and subsequently breached the AI development platform Hugging Face, prompting an investigation into other potential attacks.

OpenAI said its broader review remains ongoing and pledged transparency. Meanwhile, Anthropic disclosed three autonomous cyberattack incidents involving its models during testing in July, while Meta also released evidence of similar attacks.

Albanese announced a government task force, a referral to Parliament’s Joint Select Committee on AI and consideration of referring the matter to the Australian Federal Police.

Anthropic researcher resigns over risk of AI causing human extinction within a decade

He told the media that the technology could spiral “out of control” as early as the end of 2027.

Meanwhile, AI industry leaders warned the UN Security Council that increasingly autonomous systems could outpace human oversight, concentrate power and pose risks to humanity. Altman called for “extreme care,” urging governments and competing nations to cooperate on AI governance while avoiding both blind optimism and excessive pessimism.

“As this process becomes more automated, the pace of AI progress could accelerate rapidly,” he told the Security Council, stressing, “This moment calls for extreme care.”

Anthropic CEO Dario Amodei pledged that his company would “slow down as much as necessary” to ensure the safety of successive AI releases, acknowledging that poorly managed AI could threaten humanity.

The warnings followed a series of incidents involving autonomous AI cyberattacks and growing concerns about the industry's race to develop increasingly capable systems.

Executives and AI experts also called for stronger international oversight, safety standards and cooperation, but disagreements persist over how to regulate the technology.

Hugging Face CEO Clement Delangue argued that open-source AI is essential for defending against AI-powered attacks, while Yoshua Bengio urged governments to turn public concern into practical action.

Britain and France backed efforts to establish global AI rules, whereas the administration of US President Donald Trump opposed international regulation, with White House science adviser Michael Kratsios arguing that rapid technological progress should not be constrained and Trump dismissing international AI regulation as a “globalist scheme”.

Press TV’s website can also be accessed at the following alternate addresses:

www.presstv.co.uk