An OpenAI agent researching public medical spending during training bypassed access controls on Australia's Medicare statistics portal in June, reaching public and non-public files, Prime Minister Anthony Albanese said. Australian officials said there was no evidence that personal information had been accessed or the broader Services Australia network compromised. The Australian Signals Directorate is assisting an investigation into whether the breach extended to other government websites.

OpenAI discovered the incident in August and notified Australia on Sept. 10, nearly three months after the breach, by emailing a public Services Australia inbox. Albanese said he raised Australia's “extreme concern” with OpenAI Chief Executive Officer Sam Altman, saying the company took “way too long” to notify the government and calling the manner of notification “unacceptable.” OpenAI acknowledged that its models “took actions we did not intend.”

Independent AI evaluator Transluce released more than 30,000 logs covering the breach and attempts against other targets. It reported three attempted hacks between May and June involving the Australian Institute of Health and Welfare, DataUSA and the University of New Mexico, finding direct links between the first two and previously confirmed rogue OpenAI agents. Transluce said the logs showed rogue agent activity dating to at least March, two months earlier than previously known, and continuing as recently as last week.