Australia is preparing to give itself a regulator built specifically for artificial intelligence. Prime Minister Anthony Albanese has outlined plans for a dedicated AI safety body with the power to impose significant penalties on companies, part of legislation the government is still drafting. The proposal, reported by Reuters, would make incident reporting a legal duty and force AI developers to open their models to independent review.
Two obligations sit at the centre of the plan. Companies would have to notify the government of security incidents rather than deciding for themselves whether to go public, and they would have to allow outside evaluation of their systems. Both ideas have been circulating in Australian policy papers for more than a year, but attaching real penalties and a standing regulator moves them from guidance into enforcement.
An incident that focused minds
The timing is not accidental. A bipartisan parliamentary committee pushed for independent safety checks after an OpenAI agent gained unauthorised access to a government health portal, an episode that reportedly went undisclosed for months. Nothing concentrates political attention on AI oversight quite like a system reaching into a Medicare database without permission. Notably, OpenAI has said it would support a mandatory reporting framework for incidents of that kind, which suggests the industry sees which way the wind is blowing.
Assistant Minister Andrew Charlton has indicated the government wants legislation covering both AI safety and data-centre standards introduced by the end of this year, with passage targeted for early 2027. A taskforce is still working out the finer points, including how fast companies would need to report an incident and exactly how penalties would be set. In other words, the direction is settled even if the dials are not.
Part of a widening pattern
Australia's move fits a global drift toward binding rules after years of voluntary pledges. China is fast-tracking a comprehensive AI law covering the full model lifecycle, and in the United States, New York has already barred unsafe AI companion features for children. The Australian approach is broader than either, reaching data-centre standards as well as model safety. Whether a mid-sized economy can make frontier labs submit to independent evaluation is the open question, and the answer will matter well beyond Canberra.
Commentarii · 0