The Albanese government will impose tough new rules on the use of AI in automated decision-making across federal departments and agencies, with the attorney general leading development of the framework and a parallel push for a digital duty of care that would put the onus on AI companies to build in safety features.
What is automated AI decision-making? It’s when a government system — welfare eligibility, tax assessment, visa processing, child support calculations — uses algorithms to make or recommend decisions that affect citizens, often with minimal human review. Australia’s illegal robodebt scheme was the canonical failure of this approach: an automated debt-recovery system wrongly accused hundreds of thousands of people of owing money, drove some to suicide, and was eventually ruled unlawful.
🔍 THE BOTTOM LINE
Australia is writing rules to stop government AI from making decisions without fairness, accuracy, and transparency safeguards — and it’s doing so in the direct shadow of robodebt. The attorney general leads the government-side rules; the communications minister leads a digital duty of care that would force AI companies to proactively address harm. Both are in development, neither is yet legislated.
What the New Plan Covers
According to The Guardian, the national plan has four elements:
-
Government AI decision-making rules — led by Attorney General Michelle Rowland, covering fairness, accuracy, and transparency in federal agencies’ automated systems. Federal departments still use automated decision-making in some service delivery functions, and the government wants safety “built into” AI processes rather than bolted on after harm.
-
Digital duty of care legislation — led by Communications Minister Anika Wells, designed to put the onus on AI companies (Anthropic, OpenAI, Microsoft, Google — all eyeing major Australian developments) to build in safety features and proactively address potential harm. This mirrors the UK’s Online Safety Act approach but applied to AI providers rather than social platforms.
-
Privacy law reform — a second round of changes to “responsibly strengthen, modernise and simplify” Australia’s personal data protection laws, including new rules for personal data protection in AI contexts.
-
AI safety in the workplace — a new work plan element focused on how AI affects worker safety, an area where Australia’s existing regulation gap has been documented for over a year.
Why Robodebt Is the Ghost in the Room
The robodebt scheme wasn’t AI in the modern sense — it was a crude automated debt-calculation system that averaged income across reporting periods and issued debt notices without human verification. But it’s the reference point every Australian policymaker carries into any conversation about automated government decision-making.
The royal commission into robodebt found the scheme was unlawful, caused significant harm, and was enabled by a culture that treated automated systems as infallible. The new AI rules are explicitly designed to prevent the next generation of that culture — where an AI system’s output is treated as ground truth because it came from a machine.
This is why the framework prioritises “fairness, accuracy and transparency” — the three things robodebt lacked. A system can be automated and fair, but only if someone is checking the outputs and the affected person can see how the decision was made.
The Datacentre Problem Running in Parallel
Prime Minister Anthony Albanese announced a new Office of AI within his department last week, alongside plans for rules around large energy-intensive datacentres. Assistant Minister Andrew Charlton told ABC TV that Australia already has more than 200 datacentres consuming electricity equivalent to about 2% of the national grid, with projections that could triple over the next decade.
The government is implementing what it calls a “world-first AI datacentre standard” — covering where facilities can be built, proximity to housing and schools, power price impacts, and drinking water consumption. This builds on our earlier coverage of Albanese’s AI speech and the social licence question and the copyright protection commitments.
The Opposition Says It’s Not Enough
Deputy Liberal leader Jane Hume told Sky News the government’s approach was insufficient and described Albanese’s speech as lacking detail. “Talking about regulating AI by introducing rules for datacentres is a little bit like saying, ‘Well, I’ve regulated the fishing industry by setting up a licensing regime for building boats,’” she said.
She also warned against overregulation: “It’s not just about setting up certainty, it’s also about alternative opportunities.” That tension — between safety and competitiveness — is the same one that drove Australia’s earlier AI regulation retreat under Husic when Trump-era pressure made ambitious AI regulation look like an economic liability.
NZ Angle
New Zealand watches Australian AI regulation closely because the trans-Tasman regulatory gap cuts both ways. When Australia moves first, NZ companies operating across both markets face compliance asymmetry. When NZ moves first, Australian firms export their compliance burden here.
Right now, NZ has no equivalent of Australia’s planned digital duty of care, no government-specific AI decision-making rules, and no datacentre standard. The NZ government’s AI guidance remains voluntary. If Australia legislates a duty of care on AI providers, those same providers (OpenAI, Anthropic, Google, Microsoft) will face different rules on either side of the Tasman — and the default will be to apply the stricter Australian standard commercially, even where NZ law doesn’t require it.
The robodebt lesson applies directly: NZ’s Ministry of Social Development uses automated decision-making for benefit calculations. Without mandatory transparency and review safeguards, the same failure mode is structurally possible.
❓ FAQ
Is Australia banning government AI? No. The rules cover automated decision-making — the point where an AI system’s output directly affects a citizen (benefits, visas, tax). Internal AI tools for analysis or drafting are not the target. The framework is about safeguards, not prohibition.
When does this become law? The attorney general’s office is leading development; legislation is expected to follow the digital duty of care bill, which is still being drafted. The datacentre rules are expected to be legislated from next year.
What’s the digital duty of care? It’s a legal obligation on AI companies to proactively identify and address potential harms from their products — similar to how the UK’s Online Safety Act treats social media platforms. Instead of waiting for harm and suing, the duty requires companies to build safety in.
Does this affect the AI tools I use at work? The duty of care targets AI providers (the companies building the models), not end users. But if your employer deploys AI for HR, hiring, or performance decisions, the workplace safety element of the plan may eventually apply.
🔍 THE BOTTOM LINE
Australia is writing the rules robodebt should have had. The framework covers government decision-making, a duty of care on AI providers, privacy reform, and workplace safety. None of it is legislated yet — and the opposition is already arguing it’s both too much and not enough. The test will be whether the rules survive contact with the AI lobby.
