Australian Prime Minister Anthony Albanese has accused OpenAI of allowing an artificial intelligence model to breach an Australian government health statistics website, calling the incident “obviously unacceptable” and criticising the company for its delayed notification.
Albanese said the AI tool attempted to access an Australian health statistics portal in June during an OpenAI training exercise and bypassed restrictions after it was denied access to certain information.
“Today, I spoke with the CEO of OpenAI, Sam Altman, to express Australia’s extreme concern about this incident,” Albanese told reporters in New York on Wednesday.
He also criticised the company for taking too long to inform Australian authorities, saying the government was not notified until September 10.
AI model accessed non-public files
According to Albanese, the AI system accessed both public and non-public files hosted on an older government health statistics website.
The prime minister said there was no evidence that personal information had been accessed and that other government services had not been compromised.
“Nonetheless, this situation is obviously unacceptable,” he said.
The incident occurred while OpenAI was conducting training exercises designed to assess the performance of its AI models. Government Services Minister Katy Gallagher said the model had been asked to search the internet for information about Australian government spending on medicines.
OpenAI said it did not identify the unusual activity until August, when the company reviewed the model’s actions.
The company later notified the Australian government by email on September 10, sending the alert to a generic government mailbox that is checked once a day.
OpenAI agents attacked RubyGems before Hugging Face incident, researchers say
Australia launches review
Gallagher said the government was concerned about the delay in receiving the notification and the fact that it was sent to a general-purpose email address.
Defence Minister Richard Marles described the incident as an example of an AI system continuing to pursue information after being denied access.
“It asked a question, the information was not given and rather than leaving at that point, it scaled the fence,” Marles said.
Australia has launched a rapid review of the incident, involving the country’s national intelligence agency responsible for cybersecurity.
OpenAI said its investigation found that several Australian government websites and services had been accessed during an internal evaluation as its models attempted to obtain answers and statistics about Australia.
“During the course of that, our models took actions we did not intend,” the company said.
— Anthony Albanese (@AlboMP) September 23, 2026
Growing concerns over AI cybersecurity
The incident comes amid growing concerns about the ability of advanced AI systems to interact with real-world computer networks and potentially bypass safeguards.
OpenAI and other technology companies have recently faced scrutiny over similar incidents involving AI models during testing.
OpenAI previously said two of its models escaped a closed testing environment and accessed internal systems belonging to Hugging Face, a platform widely used by AI developers to store and share code.
Anthropic has also reported that its models gained unauthorised access to systems belonging to three unidentified organisations during testing.
Google said last week that its Gemini consumer AI model had hacked multiple systems by guessing login credentials.
More than 100 organisations, including OpenAI and Anthropic, signed an open letter last month calling for stronger global cyber defences against AI-powered threats.
OpenAI CEO Sam Altman and other technology executives also addressed a special United Nations Security Council meeting on Wednesday focused on the risks associated with artificial intelligence.
