Australia is putting together one of the more detailed national packages for managing AI outside of Europe. It is not a single law. It is several streams of reform running at once, each covering a different slice of the problem.
The most politically loaded piece is the one on government decisions. Australian federal departments already use automated systems to handle things like social services payments, visa assessments, aged care, and biosecurity. The new rules, led by attorney general Michelle Rowland, are about drawing clearer lines around when and how those systems can replace human judgment.
The urgency traces back to Robodebt. Between 2016 and 2019, the government ran an automated system that matched welfare payment records with tax office income data to identify overpayments. The algorithm assumed income arrived evenly across every fortnight of the year, which is not how most casual or part-time workers actually get paid. The result was that the system generated debt notices that were, in many cases, mathematically wrong and legally groundless.
Nearly 470,000 Australians received letters accusing them of owing the government money. The government eventually repaid over $750 million in wrongly recovered debts and paid $112 million in compensation to roughly 400,000 people. The class action settlement reached $1.8 billion. A Royal Commission called it "a costly failure of public administration, in both human and economic terms."
The specific flaw matters. The algorithm was built around a profile of someone in stable, full-time work, which described only about 7% of welfare recipients. Everyone else got misclassified. And when recipients did not respond to the debt notices, the system treated silence as confirmation of guilt. No human reviewed whether any of this was accurate before the letters went out.
What Australia is now trying to build is a legal structure that prevents a repeat. The proposed rules centre on transparency: agencies would need to disclose which decisions are made with automated tools, what information those tools use, and how someone can challenge the outcome. A December 2025 policy update already requires all federal agencies to assign clear accountability for each AI system they operate.
The second reform stream is the digital duty of care. This one is aimed at tech platforms rather than government agencies. The legislation, still being drafted, would legally require platforms to identify risks of harm in their services and take steps to address them before harm occurs, rather than waiting for users to report problems. It is a meaningful shift: currently, platforms mostly act after complaints. Under the new model, they would need to act before.
The third stream is the data centre question, and this is where the numbers get serious. Australia's data centres already consume about 2% of the national electricity grid. By 2030, that figure is projected to nearly triple. By the mid-2030s, some forecasts put data centre demand at 9% to 11% of total national electricity use. Australia ranked second globally in 2024 as the most attractive destination for data centre investment, after only the United States.
The government's response is a set of formal expectations, published in March 2026, that tie new data centre approvals to energy, water, and community impact commitments. Operators must bring their own renewable energy supply, cover the full cost of grid connections, and avoid drawing on drinking water without clear justification. Projects that do not align with those expectations will not get priority treatment in regulatory approvals. That is a meaningful lever: slower approvals mean delayed revenue.
For businesses that rely on AI tools, none of this changes anything in the next few months. But it signals where costs are heading. If you are using cloud services or AI platforms that run infrastructure in Australia, the cost of operating that infrastructure is going up. That eventually shows up in pricing.
For businesses that handle customer data and make decisions using automated scoring, flagging, or sorting, the regulatory direction is toward more explainability. The question regulators are starting to ask is: can you show a person why your system made the decision it made about them? If the answer is no, that is the gap to close now.