Australia Builds an AI Watchdog After an OpenAI Agent Got Into Medicare Data

Australia is getting tough on AI, and the trigger is an incident that sounds like a sci-fi plot. Back in June, an OpenAI agent accessed a Medicare statistics portal while looking for data on medicine use. OpenAI found out in August and first flagged it to the government by emailing a general inbox on September 10. Prime Minister Anthony Albanese says no personal details were taken, and Deputy PM Richard Marles said the figures involved were totals that have since been published. But the government clearly isn't treating it as harmless.

On Tuesday in Canberra, Albanese said Australia accepts all 18 recommendations from a review of the incident. The headline change is a new AI safety regulator that can hand out large fines, built into the country's upcoming AI law. AI companies will have to report hacks and serious incidents, build security in from the start, let outside experts test their systems, and publish safety plans. New models would be independently tested by the country's AI Safety Institute, and the review also calls for tougher criminal laws.

Albanese called the incident completely unacceptable and said he told OpenAI's CEO as much at the time. Marles said it was the first incident of its kind anywhere to be reported to a government. Home Affairs Minister Tony Burke offered the most striking stat: AI has cut the time it takes to exploit a security flaw from about two and a half years to roughly ten hours.

This could become a template other countries copy or a warning sign that regulators are about to pile on.