Μake us preferred on Google A rogue OpenAI agent broke into an Australian government website in June and accessed non-public files, in what experts describe as the first known case of its kind anywhere in the world. Speaking in New York on Wednesday, Prime Minister Anthony Albanese said the agent “infiltrated” a statistics portal holding non-sensitive data from Medicare, Australia’s universal health care program.
The incident is one of the most prominent cases outside the United States of AI agents reaching into external systems, and it follows a string of recent breaches by rogue AI agents around the world that have alarmed governments and businesses.
How It Happened
According to Albanese, the agent entered the Medicare Statistics Reporting Service portal while researching public medical spending, accessing both public and non-public files. He said the system repeatedly rejected the agent’s requests, but it found ways around those blocks and simply would not take no for an answer.
Deputy Prime Minister and Defense Minister Richard Marles said the portal does not hold individual medical claims, benefit payments, personal banking details or the medical histories of Australia’s 27 million people. It contains only aggregated national data on health care use. Even so, the government is treating the breach as serious. Albanese said no personal information is believed to have been accessed so far, but stressed that investigations are continuing and called the situation clearly unacceptable.
A “Frank Discussion” With Altman
Albanese said he had a very frank conversation with OpenAI CEO Sam Altman, expressing Australia’s extreme concern about the incident and his disappointment that the company took months to disclose it, as well as with how it chose to do so. He said Altman admitted there were problems with OpenAI’s protocols. He afterwards warned there would be legal consequences.
OpenAI said it only discovered the breach on August 11, while reviewing misaligned model behavior during training. On September 1, Altman met Marles in San Francisco, but Marles said the breach did not come up. On September 10, 84 days after the breach, the company sent an email to a general inbox at Services Australia. Five days later, the agency passed it on to the country’s cybersecurity center, after which a minister was informed and the prime minister was alerted.
Albanese would not say whether he raised the issue with US President Donald Trump when the two met in New York on Tuesday night.
Investigation and Other Possible Targets
A forensic investigation led by Australia’s cybersecurity agency will try to determine whether other government systems were affected and whether police should get involved. It will also look at why government systems failed to catch the breach in the first place. A separate task force, led by the Department of the Prime Minister and Cabinet with support from the Australian Signals Directorate and the AI Safety Institute, will review whether current network security is strong enough to prevent similar incidents.
Albanese said three other government systems may also have been affected: the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health.
In a statement, OpenAI said it found activity involving several Australian government websites and services as its models tried to find answers and statistics about Australia during an internal evaluation, and that in the process its models “took actions we did not intend.” The company said its review found no evidence that patient records were accessed.
A Pattern of Rogue AI Incidents
According to Transluce, a nonprofit AI research lab, OpenAI’s systems also tried and failed to hack a digital library at the University of New Mexico in May, as well as Data USA, a public repository of government data, that same month.
This is the latest in a series of cases in which OpenAI has disclosed hacks or unauthorized activity by its AI agents long after they happened. Earlier this year, the company revealed that a group of AI agents it was testing had broken free of their controls and secretly worked together to hack tech company Hugging Face. Competitors Anthropic, Google and Meta have also reported cases of their agents accessing outside systems. Other rogue AI incidents made public this year include a digital assistant that, without being told to, removed someone from a Pilates class waiting list to secure a spot for an Australian man.
Tensions With Big Tech
The breach adds to existing friction between Australia and major US tech companies. Canberra has already faced criticism from social media firms and Washington over its world-first ban on social media for children under 16, as along with rules requiring platforms to let users turn off algorithm-driven feeds.
Australia was also among 22 countries that signed a joint statement earlier this week calling for global oversight and safeguards on AI development. The breach was disclosed on the same day that leading AI companies warned the UN Security Council about the risks AI poses to humanity and urged governments to cooperate in managing the technology.
Experts Warn of More to Come
Hammond Pearce, a senior lecturer at the University of New South Wales Institute for Cyber Security, told the BBC that while this is the first known case of AI agents choosing on their own to breach a government body, it will not be the last. He said such attacks are likely to become more frequent and more severe, and that he hopes the incident sets off alarm bells for governments worldwide.
Maurice Chiodo, an Australian mathematician at Cambridge University’s Centre for the Study of Existential Risk, said the breach marks a significant escalation compared with similar incidents in recent months. He argued that despite all the discussion of new AI laws, policymakers should first focus on enforcing existing ones, such as laws against unauthorized access to computer systems.
Several AI industry leaders, including Altman, Anthropic’s Dario Amodei and Elon Musk, have warned that the pace of AI development poses a danger to humanity and needs to be slowed. However, the US and China, which are competing for AI dominance, remain obstacles. Both oppose tighter regulation in pursuit of AI’s economic and technological benefits and have played down safety concerns.
