Australia to Clamp Down on Government Use of AI in Automated Decisions
Ns applying for government support or dealing with a federal agency could soon carry a new guarantee: a machine will not have the final word
Australians applying for government support or dealing with a federal agency could soon carry a new guarantee: a machine will not have the final word on decisions that shape their lives. The Albanese government on Sunday set out plans for tough rules governing the use of artificial intelligence in automated decision-making across the public sector, the centrepiece of a new national AI plan.
Attorney-General Michelle Rowland will lead the drafting of the framework, which ministers want built around three principles — fairness, accuracy and transparency. The protections are expected to stretch beyond government itself, reaching into consumer rights, workplace safety and privacy, and the package lands alongside a separate Labor push for digital duty of care legislation aimed at the broader tech sector.
What changes inside government
Federal departments and agencies still lean on automated systems for parts of their service delivery, and that reliance has quietly grown. Officials and ministers now worry that increasingly capable AI could hollow out human-led processes altogether unless guardrails go up first. The new rules are meant to force safety into the design of those systems before that happens, not after.
The timing is not accidental. Australia is in the grip of a datacentre construction boom, AI adoption is accelerating across the economy, and global giants including Anthropic are looming ever larger in the policy debate. Prime Minister Anthony Albanese and Labor MP Andrew Charlton have both put their names behind the national plan, signalling that Canberra wants the economic upside of AI — but on terms the government sets.
The long shadow of robodebt
Any Australian debate about automation runs straight into one word: robodebt. The illegal automated debt-recovery scheme, which wrongly pursued welfare recipients en masse, became a national lesson in how badly things go wrong when algorithms replace judgment. Its failure exposed just how unreliable automated processes can be when the people affected have no meaningful way to challenge them.
That history hangs over everything Rowland now drafts. Critics will measure the new framework against a simple test: could it have stopped robodebt before it started? Ministers clearly know it. The emphasis on human oversight and transparency reads like a direct answer to that scandal, even if officials are careful not to frame it that way.
A debate stretching well beyond Canberra
Australia is not moving alone. Three days earlier, on the other side of the Pacific, Wyoming lawmakers sparred over a draft bill that would stop AI systems, large language models and similar automated tools from making what it calls "consequential decisions" entirely on their own — defined as decisions affecting "a person's rights, benefits or obligations provided by law." The proposal would also force agencies to keep records of which employees use such systems.
Staff who drafted the Wyoming bill openly borrowed from efforts in New Jersey, Colorado and Illinois, and its definition of AI — any machine-based system that infers from inputs how to generate outputs capable of influencing physical or virtual environments — echoes language now circulating in legislatures worldwide. The details differ. The instinct is the same: keep a human in the loop wherever rights, benefits or obligations are at stake.
For ordinary people, the shift is concrete rather than abstract. A disputed benefit, a flagged claim, a fine issued in error — under the emerging rules on both continents, someone with a name and a desk would have to answer for those calls.
Rowland's drafting work now begins in earnest, with the digital duty of care legislation expected to follow close behind. Departments running automated systems have been put on notice, and the first real test will come when the draft rules land and agencies discover exactly how much of their machinery needs rebuilding.
