Australia is moving to stop government algorithms from making consequential decisions about citizens without oversight. The Albanese government has begun drafting a national plan that would subject AI used in automated decision-making by federal departments and agencies to new rules covering fairness, accuracy, and transparency. The push arrives alongside proposed digital duty of care legislation and a broader plan to stand up an office of AI inside the prime minister's department, as the country grapples with rapid adoption and a datacentre construction boom.
This is not a single law. It is a bundle of reforms that together change how both government and AI companies operate on Australian soil. The automated decision-making rules target the quiet ways agencies already lean on models for service delivery. The digital duty of care puts the onus on AI firms to build safety in from the start. Separately, the government is progressing privacy law reforms and workplace AI safety rules, plus world-first standards for energy-hungry datacentres.
Why automated decision-making is the flashpoint
The specific worry is that federal agencies already use automated systems in some service delivery functions, and that those systems could edge out more human-led processes. When an algorithm screens a welfare claim, triages a visa application, or flags a tax position, the people affected rarely see the logic or get a meaningful appeal. Australia is not starting from zero here. It has existing guidelines, but the new plan would turn soft guidance into enforceable expectations around fairness, accuracy, and transparency for decisions that touch lives.
The lesson Australia is absorbing came from elsewhere. Other governments have discovered that once an automated pathway exists, agencies default to it because it is cheaper and faster, even when the stakes are high. Writing rules now, before the systems are entrenched, is the cheaper path to keeping a human in the loop where it matters. The assistant minister for science, Andrew Charlton, framed the pace of change as already challenging, noting Australia has more than 200 datacentres consuming about 2% of the national grid, with projections that could triple within a decade.
A digital duty of care flips the default
The digital duty of care legislation, led by communications minister Anika Wells, is the part that should worry AI vendors most. Instead of regulators chasing harms after the fact, the bill would put the responsibility on AI companies to build in safety features and proactively address potential harm. That mirrors a shift seen in other jurisdictions, where the burden moves from users proving damage to builders proving they took reasonable steps to prevent it.
For startups shipping models into Australia, this changes the product checklist. Documentation, evaluation, and incident response stop being optional polish and become part of the compliance surface. A model that is accurate in a lab but unpredictable in a high-stakes government workflow is exactly the kind of deployment the duty of care is built to catch. The practical effect is that safety work has to happen before launch, not after a scandal.
An office of AI and a whole-of-government posture
Prime Minister Anthony Albanese used a major speech last week to announce a new office of AI within his department, part of a whole-of-government approach to managing the technology. The framing matters. Rather than scattering responsibility across agencies, the government is creating a central node that can set standards, coordinate procurement, and speak with one voice on AI policy. That centralization makes it easier for vendors to know the rules, but also easier for the government to enforce them consistently.
The plan also bundles privacy reform, billed as an effort to modernize Australia's personal data protection laws, and a workplace AI safety focus. Taken together, the package signals that Australia intends to treat AI as infrastructure, not just software. The datacentre standards, expected to be legislated from next year, will govern where massive facilities can be built and how they affect local communities, power prices, and water use.
What this means for AI vendors and the global regulatory race
Australia's move is a signal that the regulatory center of gravity is shifting from voluntary pledges to enforceable duties, and vendors should plan accordingly. First, the duty of care model rewards companies that can show their safety work rather than merely claim it. Builders who treat evaluation, logging, and red-teaming as first-class product features will clear these regimes faster than those who bolt them on late. The cost of compliance is lowest when it is designed in, not retrofitted.
Second, government procurement is about to become a powerful lever. Once federal agencies must use AI that meets fairness, accuracy, and transparency standards, the buying power of the state pulls the whole local market toward compliant tools. Enterprises that sell into government will need the same evidence packs, and that demand cascades into the private sector as a de facto standard.
Third, this is a template other middle-power economies will watch. Australia is positioning itself as a rule-setter on AI infrastructure, from datacentre siting to model duty of care, rather than a follower of either the US light-touch approach or the EU's sweeping rulebook. For founders, that means the next few years of Australian policy will be a live experiment in how a resource-rich democracy balances AI growth against citizen protection. The companies that engage with that process early, through consultation and transparent deployment, will have an easier time as the rules harden.