While the loudest arguments about AI concern frontier models and who gets to regulate them, two governments spent the past week worrying about something more immediate: what the technology is already doing to ordinary people. Australia and California both moved to put guardrails around it. What is striking is that they are guarding against opposite dangers. One government fears the machine that decides. The other fears the machine that replaces.
In Canberra, the Albanese government published a list of AI priorities on Sunday promising to "better regulate" the use of automated systems inside federal agencies like Centrelink and Services Australia, and to fold in stronger privacy protections. The subtext is a national scar. Last year the Commonwealth Ombudsman found that almost a thousand jobseekers had their income support unlawfully terminated by automatic cancellations, an echo of the Robodebt scandal that has made Australians deeply wary of letting software make calls about welfare. The government is now facing scrutiny over automated elements of aged care support and an algorithm that generates budgets for disability scheme participants, which the Ombudsman is investigating. In May, the employment minister felt compelled to promise that humans would stay in the loop on unemployment payments. The fear here is precise: a wrong answer from an unaccountable system, delivered to someone who cannot appeal it.
In Sacramento, the worry runs the other way. An executive order signed by Governor Gavin Newsom in May, drawing fresh attention this week, is aimed not at AI making bad decisions but at AI making people redundant. It directs state agencies, universities and labour economists to build an early-warning system for AI-driven job losses and to prepare the cushions for when they come: severance standards, expanded unemployment insurance, retraining, even experiments with worker ownership and "universal basic capital." Agencies have 180 days to recommend updates to California's layoff-notification law so it can flag disruption sooner. With 33 of the world's top 50 private AI companies headquartered in the state, California is preparing for the possibility that the industry it hosts will hollow out the workforce that surrounds it.
The two responses share a lineage. Both come from centre-left governments that instinctively reach for the safety net, and both frame AI as something happening to workers rather than something workers choose. Prime Minister Anthony Albanese put it bluntly: "artificial intelligence is happening," and people have an interest in shaping the change "because if not, then they don't get a say." That is a very different register from the frontier-safety debate in San Francisco, where the question is whether a future model might be dangerous. In Adelaide this week, AI is set to dominate Australia's Labor conference, with the fight centred on workers' rights and a pledge to hold AI companies to Australian "values and laws."
Put side by side, the two agendas map the two ways AI actually touches most lives. It can be the invisible clerk that cuts off your payment with no one to call, or the quiet force that erases your job before anyone names the trend. Neither government is trying to steer the technology's ceiling. Both are trying to catch the people underneath it. That is a less glamorous kind of AI policy than drafting rules for superintelligence, and, for now, a far more concrete one.