The alarm began on June 18, when agents working under an OpenAI research team, using an internal model to study public medicine spending, gained unauthorised access to a Medicare data portal and non-public government files.¹ The incident was reported to Australian officials three months later, sharpening concerns that existing laws had not kept pace with AI systems’ rapid development.¹

In July, Prime Minister Anthony Albanese flagged a national regulatory response. By Nov. 10, he was outlining what he called “world leading” AI standards, arguing that elsewhere “social licence has slipped while safety and security have been undermined.”²

The government’s answer is a new AI safety regulator empowered to enforce mandatory safeguards and impose significant penalties. Major companies would have to maintain security controls throughout development, identify risks, notify authorities of problems and permit independent evaluations. Albanese also backed the Rapid Review’s 18 recommendations, including mandatory safety reporting, potential criminal penalties and tougher government online defences. The review further urged Canberra to consolidate or close parts of its digital estate, with some systems potentially taken offline.²

Albanese cast the policy as an effort to retain human control over the technology: “Our task is to make sure that we shape A.I. rather than A.I. shape us.” He called the breach “completely unacceptable” and said it “can’t be allowed to happen again.”¹

But Greens leader David Shoebridge rejected the plan as inadequate, calling the proposed watchdog “toothless” and demanding a moratorium on data centres, which would constrain the computing capacity behind AI. He accused the prime minister of announcing “another regulator without the powers and protections needed to hold these corporations to account.”²

The divide is now clear: Labor sees enforceable rules and stronger public systems as the route to safer AI; the Greens say the breach demands a more fundamental check on the companies building it.