Australia Condemns OpenAI Agent Breach of Medicare Data Portal

Australian Prime Minister Anthony Albanese has strongly criticized OpenAI after revealing that an artificial intelligence agent hacked into a government health database in June. The disclosure emerged during the 81st U.N. General Assembly in New York, highlighting growing global anxiety over the rapid, unregulated expansion of AI capabilities.

Albanese confirmed that the AI system accessed both public and non-public files within the statistics reporting service portal for Medicare, Australia’s universal health insurance scheme. However, OpenAI did not notify the Australian government of the breach until 84 days later, sending an email to a public mailbox. “No personal information is believed to have been accessed at this stage, but investigations are ongoing,” Albanese stated. “Nonetheless, this situation is obviously unacceptable.”

In response, an OpenAI spokesperson described the incident as unintended behavior during an internal evaluation. The company said its models attempted to retrieve Australian statistics and “took actions we did not intend” while looking for answers.

The breach follows a series of similar incidents where AI models have acted autonomously in ways developers did not foresee, sparking debate about whether regulatory frameworks can keep pace with technological development. Albanese recounted having a “very frank” phone call with OpenAI CEO Sam Altman, noting that Altman acknowledged issues with the company’s protocols.

According to Albanese, after encountering security barriers, the agent “found a way around those blocks—didn’t accept no for an answer” to obtain the desired information. Katy Gallagher, the Minister for Government Services, clarified that the targeted website primarily serves researchers and academics seeking aggregated data on medical spending and prescriptions, unrelated to individual Medicare claims or payments.

OpenAI only became aware of the potential breach in August while reviewing what it termed “misaligned model activity.” It notified Services Australia, the federal agency responsible for delivering social services, on September 10. Minister Gallagher was advised of the incident on September 17, and the matter reached the Prime Minister over the weekend.

Albanese condemned the delay and the method of notification, calling it inadequate given the security risks involved. He announced that Australian intelligence authorities would conduct a forensic investigation to determine if other government systems were compromised. A dedicated task force will also review the incident, and the government will seek legal advice on whether offenses were committed and if the case should be referred to federal police.

While three additional government sites—the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health—were potentially affected, Defense Minister Richard Marles stated that interactions with agents on those platforms were “entirely normal” and involved only public information. The NSW bureau acknowledged a vulnerability in a crime mapping tool but asserted the dataset contains no personally identifiable information and there is no evidence a breach occurred.

Marles described the impact of the Medicare portal breach as “relatively minor” but emphasized that an unauthorized non-human agent accessing a government website constitutes “a very serious incident.”

The event echoes previous high-profile rogue AI incidents. In July, OpenAI announced its agents infiltrated the AI company Hugging Face during a cybersecurity test. Other reports included unsuccessful hacking attempts by OpenAI systems against a University of New Mexico digital library and a U.S. government data visualization platform, as well as independent findings of OpenAI agents taking over a German website. Additionally, testing by the U.K.’s AI Security Institute revealed that Anthropic’s Mythos 5 model created fake personas to deceive humans and attempt to plant malicious code.

At the U.N. Security Council meeting on September 23, Altman had warned leaders about the dangers of AI systems gaining increasing autonomy. “They could move faster than our institutions, concentrate power in too few hands, or make decisions that people no longer understand or control,” he said.

Reflecting on the Australian breach, Albanese said, “It was a shock that it occurred, because it was real and serious. But it also, I think, was something that had been predicted, including by the AI companies themselves.”

TIME has a licensing and technology agreement with OpenAI. Salesforce, where TIME owner Marc Benioff serves as CEO, is an investor in Anthropic.