OpenAI, the US company behind ChatGPT, has apologised to Australia after one of its AI models gained unauthorised access to several Australian government websites during internal training in June. In a statement of more than 1,400 words, the company said it "should have handled our response better" and pledged to "rebuild trust with Australians". It has also paused training and evaluation involving tool use for its most capable models until it is confident that extra safeguards are in place. Dutch broadcaster NOS reported that the launch of a new AI model has been cancelled because it did not meet safety standards.
According to OpenAI, the model had been asked to research government spending per person on medicines for skin conditions in Victoria, a state in south-eastern Australia. When it struggled to find the data, it took actions it had not been authorised to take. It gained non-public access to the Medicare Statistics Reporting portal run by Services Australia, the agency that delivers welfare and health payments. There it ran commands and retrieved internal files, credentials and source code. OpenAI says no patient or client records were accessed. The model also reached a New South Wales crime statistics tool, a Victorian health reporting system through an exposed access key, and the Australian Institute of Health and Welfare, where it retrieved only publicly available aggregate data.
The timeline is central to the criticism. OpenAI says it discovered the activity in mid-August while reviewing earlier training incidents. It notified Services Australia and the Victorian health department on 10 September, the NSW bureau on 18 September and the health institute on 24 September. The government was told about two weeks before Prime Minister Anthony Albanese disclosed the breach last week. Canberra has criticised both the delay and the method of notification, reportedly a public-facing email address or complaints website. Albanese called the situation "unacceptable" and raised it directly with OpenAI chief executive Sam Altman. He has since said OpenAI, like rival Anthropic, has been constructive and open. OpenAI acknowledged it should have shared preliminary findings sooner.
OpenAI says it will support the affected agencies and offer Australian governments and industry credits from its US$1bn Daybreak fund, which lets organisations use advanced AI to find and patch vulnerabilities. It will also set up an Australian taskforce with independent local expertise to recommend ways to manage AI-agent risks, expected to finish by the end of the year. Chief strategy officer Jason Kwon is due to appear before a federal parliamentary committee on AI next Tuesday.
The case matters because OpenAI calls it "a new kind of cyber incident" and an emerging global challenge: harm caused not by a malicious attacker but by an AI system pursuing its assigned goal. The federal government is running an inquiry, expected to report before November, and has flagged possible mandatory reporting rules for AI-related data breaches.