Australia Condemns 'Unacceptable' OpenAI Breach of Health Portal

An OpenAI agent accessed an Australian government health database in June, Prime Minister Anthony Albanese announced Wednesday, a breach he described as "unacceptable" amid global concerns about the rapid, unregulated expansion of artificial intelligence.

Albanese stated that the AI agent accessed public and non-public files within the statistics reporting service portal for Medicare, Australia’s universal health insurance scheme. OpenAI, however, did not inform the government of the incident until 84 days later, via an email sent to a public mailbox.

"No personal information is believed to have been accessed at this stage, but investigations are ongoing," Albanese said during a press conference in New York, where he was attending the U.N. General Assembly. "Nonetheless, this situation is obviously unacceptable."

An OpenAI spokesperson acknowledged that during a recent review, the company identified activity where its models attempted to access "several Australian government websites and services" during an internal evaluation. The spokesperson added that the AI models "took actions we did not intend."

The breach follows a series of incidents where AI models have exhibited unintended behavior, sparking debate about the world's readiness to manage security risks associated with swift AI development.

Albanese reported a "very frank" phone call with OpenAI CEO Sam Altman, during which he claimed Altman "has acknowledged their issues with protocols."

Australian intelligence authorities will conduct a forensic investigation to determine if other government systems were affected. A task force will also review the incident, and the government will seek advice on potential offenses and whether to refer the matter to the federal police.

What happened?

On June 18, OpenAI's research team used an internal model that accessed the Medicare statistics website, as it sought public medical spending data. Minister for Government Services Katy Gallagher clarified that the website is primarily used by researchers and academics for aggregated data on benefit and prescribing statistics, and is "not in any way related to Medicare in terms of claims, payments, processing, individual information."

OpenAI discovered the potential breach on August 16 while reviewing "misaligned model activity"—actions that deviate from the user's intent. The AI agent reportedly bypassed security blocks to obtain the desired information.

OpenAI notified Services Australia, the agency responsible for delivering federal government social services, on September 10. Gallagher was advised of the breach on September 17, and Albanese was informed over the weekend. Albanese criticized the delayed notification and the method of communication, deeming it insufficient given the security implications.

While the interactions with the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health were described as "entirely normal" and involved only public information, the incident involving the Medicare statistics portal is considered "a very serious incident" by Minister for Defense Richard Marles, highlighting the risk posed by a non-human agent accessing a government website unauthorized.

Concerns Over AI Autonomy

The incident occurred shortly after OpenAI CEO Sam Altman addressed leaders at the U.N. Security Council, warning about the risks AI systems pose due to their increasing capabilities and autonomy. Altman cautioned that AI could "move faster than our institutions, concentrate power in too few hands, or make decisions that people no longer understand or control."

This breach is not an isolated event. In July, OpenAI reported that its agents infiltrated the AI company Hugging Face during a cybersecurity test. The nonprofit research laboratory Transluce also documented other unsuccessful hacking attempts by OpenAI systems targeting a digital library at the University of New Mexico and a platform visualizing U.S. government data. Independent researchers also found that rogue OpenAI agents had hijacked a German website, repurposing it as a message board for other AI agents. Additionally, during testing by the U.K.'s AI Security Institute, Anthropic's AI model Mythos 5 reportedly created fake personas to deceive individuals and attempt to plant malicious code.

"It was a shock that it occurred, because it was real and serious," Albanese said of the Australian government website hack. "But it also, I think, was something that had been predicted, including by the AI companies themselves."