Every government that has taken artificial intelligence seriously has had to answer the same awkward question: do you try to stop the technology, get out of its way, or attempt something in between. Australia, after several years of consultation papers, safety summits and cautious ministerial speeches, has largely settled on the third option. The verb of choice in Canberra is now “manage”, and a new analysis from Macquarie University asks what that word will actually mean once it is turned into law.
The framing matters because “managing” AI sounds reassuringly moderate, yet it hides a set of genuinely hard choices. Which uses of the technology count as high risk. Who carries the legal liability when an automated system gets a decision wrong. How much a mid-sized Australian business is expected to document about a model it merely bought off the shelf. The Macquarie piece, published as part of the university’s research commentary, works through why the tidy political language of management is harder to deliver than it looks.
How Australia arrived at ‘management’
Australia’s approach has been evolving in the open. In 2024 the federal government released its proposals paper on mandatory guardrails for AI in high-risk settings, floating obligations such as testing, transparency, human oversight and accountability for the developers and deployers of the riskiest systems. The idea was never to regulate spell-checkers and photo filters. It was to draw a line around applications where a faulty or biased model could cost someone a job, a loan, a visa or their liberty, and to require the organisations behind those systems to prove they had done their homework.
That is the essence of the “management” posture. Rather than an EU-style comprehensive AI Act that classifies the entire market from the outset, or a hands-off American instinct to let the courts and the market sort it out, Australia has leaned towards a risk-based model bolted onto laws it already has. Privacy, consumer protection, anti-discrimination and workplace rules would do much of the heavy lifting, with new guardrails filling the gaps where AI does something those older laws never anticipated.
Two ways of reading the same word
Here the viewpoints diverge sharply, and the Macquarie analysis is useful precisely because it does not pretend the debate is settled.
Industry groups and many technology firms broadly welcome a managed, principles-based approach because it promises flexibility. Their argument runs that AI is moving too fast for prescriptive rules, that a heavy regulatory hand would push investment and talent offshore, and that a light-touch regime lets Australia capture the productivity gains economists keep promising. For a country that imports most of its foundational models, the reasoning goes, the smart play is to regulate outcomes rather than the technology itself, so local firms can adopt whatever tools the global market produces without tripping over a thicket of compliance.
The counter-view, voiced by consumer advocates, unions, legal scholars and a good many academics, is that “management” can quietly become a synonym for “not much”. These critics point to the gap between the government’s stated ambitions and the slow pace of binding rules actually reaching the statute book. They worry that a system built on self-assessment lets companies grade their own homework, that voluntary standards without penalties change little, and that the people most exposed to automated decision-making, including welfare recipients, jobseekers and tenants, have the least capacity to challenge a bad outcome. The Robodebt scandal hangs over this conversation like a warning, a reminder of what happens when automated government decisions run ahead of accountability.
What it means for Australia
For Australian businesses and institutions, the practical stakes are already arriving ahead of the legislation. Banks, insurers, universities and government agencies are deploying AI into customer service, hiring, fraud detection and student assessment right now, and they are doing so without a settled national rulebook telling them where the lines sit. A managed regime, if it is clear, could give those organisations the confidence to invest, knowing what compliance looks like. If it stays vague, many will either over-invest in cautious legal cover or push ahead and hope the rules, when they come, are kind.
The uneven burden is another distinctly Australian concern. The country’s economy is dominated by small and medium enterprises that lack the compliance teams of a Commonwealth Bank or a Telstra. A guardrails regime that demands extensive documentation and risk assessment could be trivial for a large corporate and genuinely daunting for a suburban accounting firm or a regional health provider trying to use an AI scheduling tool. Getting the calibration right, so that the rules bite where harm is real without smothering ordinary adoption, is the whole game, and it is exactly the detail that the word “manage” papers over.
There is also a sovereignty dimension. Australia is a rule-taker on the underlying technology, so much of what happens here will be shaped by decisions made in California, Brussels and Beijing. Managing AI locally, in that context, is partly about ensuring that global systems behave acceptably when they touch Australian citizens, and partly about building enough domestic capability, in skills, standards bodies and public-sector expertise, that the country is not simply accepting whatever defaults arrive from overseas.
What happens next
The immediate question is whether Canberra converts its consultation work into binding law, and how tightly it defines “high risk”. A voluntary AI safety standard already exists to guide businesses, but voluntary is the operative word, and the pressure to make at least part of the regime mandatory is unlikely to fade. Expect continued argument over whether AI needs its own dedicated act or whether amendments to existing privacy and anti-discrimination laws can carry the load, and expect the tech sector to lobby hard against anything that looks like the European model.
For readers, the Macquarie University analysis is a helpful antidote to the false comfort of the word itself. “Managing” AI is not a decision that ends the debate. It is the start of a long negotiation about who bears the risk when the machines get it wrong, and that negotiation is only just beginning in Australia.
Sources: Macquarie University via GNews.
