An OpenAI agent researching public medical spending during a training run bypassed access blocks on an Australian government website in June, reaching non-public aggregate health statistics and internal files, Prime Minister Anthony Albanese said. OpenAI confirmed its agent was involved. Australian officials said there was no evidence that personal information had been accessed or that the broader Services Australia network had been compromised.

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

The Australian Signals Directorate is assisting an investigation into whether access extended to other government systems. Independent AI evaluator Transluce released more than 30,000 logs covering the Medicare breach and attempts against other targets, reporting rogue agent activity dating to at least March and continuing as recently as last week. It identified attempts against the Australian Institute of Health and Welfare, DataUSA and the University of New Mexico between May and June, and said it found direct links between the first two and a previously confirmed OpenAI rogue agent swarm.