Sydney (GNP): An artificial intelligence model developed by OpenAI breached part of an Australian government website during a training exercise, raising concerns over AI safety and cybersecurity controls.
Australian Prime Minister Anthony Albanese described the incident as “obviously unacceptable” and raised the matter directly with OpenAI chief executive Sam Altman after the company informed the government about the breach in September.
The AI model attempted to access a health statistics portal in June while being tested to assess its performance. Australian officials said the system bypassed restrictions and reached a section containing private files.
Albanese said there was no evidence that personal information had been accessed and confirmed that other government services were not affected.
He nevertheless criticised the delay in communication, noting that the government was only informed on 10 September, and that the notification had come through a general email inbox checked just once a day.
Government Services Minister Katy Gallagher said the model had been asked to search online for information about Australian medicine spending during an internal evaluation.
OpenAI said it discovered the activity during an extensive review of its AI models and found that the system had taken actions that were not intended.
Australia has since launched a rapid review of the incident, involving its national cybersecurity authorities.
Defence Minister Richard Marles described the AI system’s behaviour as unexpected, adding that it continued even after access was restricted.
Also Read: Former CDS Anil Chauhan Joins Hindu Nationalist-Aligned Think Tank
The incident comes amid wider discussions about the risks linked to advanced AI systems, with OpenAI and other technology companies previously reporting cases of AI models carrying out unauthorised actions during testing.
OpenAI, Anthropic and more than 100 other organisations have recently called for stronger cybersecurity measures to address potential threats involving artificial intelligence.
The Australian incident has renewed questions about how AI models should be tested, monitored and controlled before being deployed in real-world environments.
Officials and technology companies are now examining safeguards to ensure future AI systems operate within defined limits while their development continues.
Anthony Albanese said, “It’s obviously unacceptable, and I’ve raised my concerns directly with the company.”
