The race to build ever more capable artificial intelligence has become one of the defining contests of the decade, pitting the United States against China, and drawing in a small number of enormously well-resourced companies chasing systems that can reason, code and act with less and less human oversight. In that sprint, a growing chorus of critics argues, one thing keeps getting pushed to the back of the queue: the safety of the people who will live with the results.
That is the case put by a recent opinion piece in The Canberra Times, which argues that the competitive frenzy around AI has inverted the usual order of priorities. In the rush to be first, the argument runs, capability is celebrated and safety is deferred, with governments reluctant to slow the party while the money and the geopolitical prestige are flowing. It is a familiar tension in technology policy, but the scale and speed of the current AI build-out have given it fresh urgency.
Why the “arms race” framing matters
The phrase “arms race” is doing real work here. It signals that the incentive to move fast is not just commercial but strategic, and that no single player feels able to pause without ceding ground to a rival. That logic has shaped the behaviour of the largest developers, who have poured tens of billions of dollars into training runs and data centres while acknowledging, often in the same breath, that they do not fully understand how their most advanced models work.
The worry is not science fiction. More immediate concerns include AI systems that generate convincing disinformation, tools that can be repurposed to help with cyber attacks or bioweapon research, automated decision-making that quietly discriminates, and a wave of highly capable “agents” that take actions online with limited supervision. When the commercial reward is speed, the argument goes, the unglamorous work of testing, red-teaming and independent oversight tends to be trimmed first.
Two ways of reading the same problem
There is a genuine debate here, and it does not split neatly into believers and sceptics. On one side sit safety advocates and many researchers who argue that binding rules, independent evaluation and mandatory transparency are the only things that will hold up once commercial pressure intensifies. They point out that voluntary commitments are easy to sign and easy to walk back, and that history offers few examples of powerful industries policing themselves effectively without a regulator in the room.
On the other side are those who warn that heavy-handed regulation could push investment and talent offshore, entrench the incumbents who can afford compliance teams, and leave a country like Australia dependent on rules written in Brussels, Washington or Beijing. For this camp, the priority is capability and sovereignty: build the local expertise, the compute and the industry first, and worry that a nation which regulates what it cannot itself produce ends up with the costs of oversight and none of the benefits. Both positions can be held sincerely, and both contain a grain of truth, which is precisely what makes the policy choices so hard.
Where Australia actually stands
For all the talk of a global contest, the practical question for readers here is what any of this means on the ground. Australia has so far chosen a light-touch path. Rather than a dedicated AI Act of the kind the European Union has passed, the country has leaned on a Voluntary AI Safety Standard and a set of existing laws, from privacy and consumer protection to anti-discrimination, that were mostly written before generative AI existed. The former Albanese government ran a consultation on mandatory “guardrails” for AI in high-risk settings, floating obligations such as testing, transparency and human oversight, but binding legislation has not yet followed.
That gap is the heart of the matter. Industry and Innovation Minister Tim Ayres has inherited a portfolio in which the ambition to make Australia an AI adopter and, eventually, a producer sits uneasily beside the unfinished business of regulation. The eSafety Commissioner, Julie Inman Grant, has been among the more forward-leaning voices in the public sector on AI-enabled harms, but her remit covers online safety rather than the full sweep of model development. The result is a patchwork, and the op-ed’s central complaint is that a patchwork assembled during a boom tends to favour the builders over the public.
What it means for Australians
Australia is not a frontier AI developer, and it is unlikely to become one at the scale of the US labs. That makes the country largely a rule-taker, importing systems trained elsewhere and deploying them across banking, health, government services, mining and retail. In that position, the safety choices made overseas land here whether or not Australians had a say in them. A model that behaves unpredictably in Sydney was almost certainly trained in California, and a regulatory gap in Canberra simply means the imported risk arrives unmanaged.
There is a second, quieter Australian stake. The current wave of investment in local data centres and AI infrastructure, much of it announced in recent months, ties the economy more tightly to this technology just as the safety questions are becoming sharper. The more the country commits to AI as an engine of productivity, the more it has riding on those systems being trustworthy. Deferring the safety conversation does not make the exposure go away, it just moves the reckoning to a less convenient moment.
What happens next
The near-term signal to watch is whether the government converts its consultation on mandatory guardrails into actual legislation, and how tightly any rules are drawn around high-risk uses. Also worth watching is how Australia positions itself in the emerging web of international AI governance, from the network of national AI safety institutes to standards work happening through bodies the country already belongs to. If Canberra wants a voice in how the frontier is governed, participation now buys influence later.
The uncomfortable truth underneath the op-ed is that safety and speed are not opposites so much as competing claims on the same scarce attention and money. Whether Australia can hold both at once, rather than treating safety as the thing it will get to once the race is won, is a test the country has not yet passed. The arms race framing suggests there is no natural finish line, which means the safety work cannot wait for one.
Sources: The Canberra Times.
