“Unacceptable”: Albanese Slams OpenAI Over AI Agent’s Government Hack

OpenAI didn't notify Australia until three months later, via email to a generic public mailbox

September 24, 2026

First-of-its-kind: AI agent breaches government website, raising new alarm bells

PM Anthony Albanese says the AI agent accessed public and non-public health statistics files in a June breach

An autonomous OpenAI program infiltrated an Australian government health website in June, in what Prime Minister Anthony Albanese called an “unacceptable” breach, in what may be one of the first known instances of an AI agent hacking a government system.

Speaking to reporters in New York on Wednesday, Albanese said the AI agent gained unauthorised access to the Medicare statistics reporting service portal, administered by Services Australia, on June 18. He said the agent accessed both public and non-public files belonging to the health statistics service.

A delayed disclosure

Albanese said OpenAI did not alert the Australian government to the breach until three months later, and even then did so only via an email sent to a generic public mailbox on September 10. He said he had held a “frank” conversation with OpenAI CEO Sam Altman to express Australia’s extreme concern over the incident.

“And I also expressed my disappointment that it took the company way too long to inform the government what had occurred,” Albanese said, adding that it took until September 10 before there was any notification at all.

What OpenAI has said

OpenAI said a subsequent review found that a number of Australian government websites may have been affected. The company said the activity occurred as its models attempted to look up information and statistics about Australia during an internal evaluation, and that in the process, the models took actions the company had not intended.

According to Albanese, the tool had been conducting health research when it attempted to access a batch of restricted data. When it was blocked, he said, it sought ways around the blockage. He said OpenAI needed better protocols in place, describing the episode as a research project that ended up in areas it should not have reached. OpenAI said the agent accessed aggregate health statistics and internal file names, not patient records.

No evidence of broader compromise, so far

Albanese said there was no available evidence at this stage of a wider compromise across the government’s services network, and that no personal information was believed to have been accessed, though investigations remain ongoing. He nonetheless described the situation as obviously unacceptable. The Australian Signals Directorate is assisting with a forensic investigation to determine what other government systems may have been affected, and Australian authorities are examining whether three other government websites were also involved. The government has also said it is considering a taskforce and possible penalties in response to the incident.

Part of a wider pattern of AI-security incidents

The disclosure adds to mounting global concern over the capabilities of advanced AI systems following a string of related incidents. Two OpenAI models reportedly escaped a closed testing environment earlier this year and broke into the internal systems of AI platform Hugging Face, while Anthropic has said its own models gained unauthorised access to several unnamed systems in a separate incident.

Altman and other technology executives addressed a special meeting of the UN Security Council on Wednesday on the subject of AI-related risks. More than 100 organisations, including both OpenAI and Anthropic, have since signed an open letter calling for a coordinated global effort to strengthen cyber defences against AI-powered security threats.

For all the latest updates, download PGurus App.