A policy analysis from September 2026
Australia has spent the better part of a decade circling the same question without quite settling it: should the country write AI-specific rules before the technology matures, or build capability first and regulate what actually goes wrong. The country’s first AI Ethics Framework arrived from CSIRO’s Data61 in 2019. Its first AI Action Plan followed in June 2021, built around four objectives: transforming Australian businesses through AI adoption, attracting the world’s best AI talent, using AI to solve national challenges, and positioning Australia as a global leader in responsible AI. That 2021 plan is now explicitly archived as a document of the previous government. What replaced it, five years and one change of government later, is a far more detailed, far more operational document: the National AI Plan, released on 2 December 2025, alongside a parallel plan governing how the federal public service itself uses the technology. Between them, the two documents mark the clearest statement yet of how Canberra intends to treat AI, not as a standalone policy silo, but as infrastructure, industrial strategy, and public administration all at once.
From a Discussion Paper to a Whole-of-Economy Plan
The lineage matters here. Australia’s approach to AI policy has always run through public consultation before legislation. The 2021 Action Plan itself emerged from a 2020 discussion paper process, with submissions closing in November of that year and a call for the country to act urgently or risk falling behind the rest of the world. That same consultative instinct shaped the National AI Plan. The Department of Industry, Science and Resources describes conducting targeted consultation with industry, unions, and civil society ahead of the December 2025 release, and the plan explicitly builds on, rather than discards, the ethics-first groundwork laid in 2019.
What changed between 2021 and 2025 is the scale of ambition. The earlier Action Plan was framed around trust and adoption; the National AI Plan is framed around economic capture. The government estimates that AI and automation could generate up to $600 billion a year toward Australia’s GDP by 2030, and points to around 650 AI companies already headquartered in the country, backed by roughly $7 billion in foreign investment into Australian AI technologies over the preceding five years and $2 billion in venture capital flowing into local AI applications in 2023 alone. The National AI Plan takes what officials describe as a whole-of-economy approach, deliberately aligned with the government’s broader Future Made in Australia industrial strategy, treating AI less as a discrete technology sector and more as a capability layered across manufacturing, resources, health, and defence.
Three Pillars: Capture, Spread, Keep Safe
The National AI Plan organises its policy commitments around three pillars, each doing different work. The first, capturing the opportunities, centres on smart infrastructure: compute capacity, connectivity, and data centre buildout, positioned against Australia’s genuine advantages in stable governance, land availability, renewable energy potential, and subsea cable connectivity. Independent modelling cited alongside the plan projects Australia’s operational data centre capacity could grow from around 0.3 gigawatts in 2024–25 to somewhere between 2.2 and 3.2 gigawatts by 2035, representing 8 to 11 percent of national electricity consumption and concentrated heavily around Sydney and Melbourne.
The second pillar, spreading the benefits, is where the plan tries hardest to distinguish itself from a narrow industry-growth document. It commits to digital skills programmes, school-level AI pilots, and an explicit acknowledgement that benefits must reach regional and disadvantaged communities rather than concentrating in capital-city tech hubs. The third pillar, keeping Australians safe, is built around what the government calls proportionate regulation rather than a single omnibus AI Act. Its centrepiece is a new AI Safety Institute, backed by $29.9 million in funding and explicitly modelled on comparable bodies such as the UK’s AI Safety Institute, tasked with testing frontier models, analysing emerging threats, and working across government with regulators, security agencies, and financial supervisors to keep pace with the technology’s risks. As Minister for Industry and Innovation Tim Ayres put it when announcing the plan, the Institute is meant to function as “a capability at the heart of government” that can identify risks on the horizon and ensure they are mitigated across every part of the Australian Government, rather than leaving that job to any single regulator.
Government as First Mover: The APS AI Plan
Ten days before the National AI Plan’s release, the government published a companion document aimed entirely inward: the AI Plan for the Australian Public Service 2025, launched on 12 November 2025 and jointly implemented by the Department of Finance, the Digital Transformation Agency, and the Australian Public Service Commission. Where the National AI Plan sets direction for the economy, the APS Plan is a delivery mechanism for government itself, structured around three mutually reinforcing pillars of its own: Trust, covering transparency, ethics, and governance; People, covering capability-building and staff engagement; and Tools, covering access, infrastructure, and support. Every public servant is promised foundational training to use generative AI tools safely and effectively, and the plan explicitly commits to engaging with unions to ensure adoption is transparent and well managed rather than imposed from above, a nod to the workforce anxieties that have accompanied AI rollouts inside government departments elsewhere.
On the tools side, the plan’s most concrete deliverable is GovAI, a centralised, Australian-hosted AI infrastructure service designed to give agencies secure access to generative AI capability at low cost while explicitly preventing vendor lock-in, alongside GovAI Chat, a shared conversational tool that lets public servants draw on government data through a familiar, auditable interface. Minister for Finance Katy Gallagher described the plan as setting “a proactive stance to increase AI adoption” so government “keeps pace with community expectations as well as international peers,” language that reflects a broader theme across both documents: Australia positioning itself as neither a first-mover risk-taker nor a laggard, but as a fast, careful follower calibrated to what allied governments are already doing.
Sovereignty, Data Centres, and the Anthropic Precedent
The National AI Plan’s most consequential regulatory move landed not in December 2025 but the following July. In mid-2026, the government announced it would develop nationally consistent standards for data centres and AI training facilities, to be considered by National Cabinet in August 2026 with legislation expected early the following year. Officials describe this as the first such regime to be legislated by any government worldwide, intended to deliver a simple, consistent framework covering energy, water, safety, and other requirements for large-scale AI infrastructure, replacing what has so far been a patchwork of state-by-state planning approval processes. The government has framed the push explicitly in sovereignty terms: ensuring AI development happens locally, where it aligns with Australia’s national priorities and economic interests, positions the country to help shape ethical standards and secure technologies rather than simply importing them.
That framing was tested almost immediately. In March 2026, the government published its Expectations of data centres and AI infrastructure developers, a document setting out the terms under which foreign AI companies should operate inside Australia. The first company to formally sign onto those terms was Anthropic, which entered a memorandum of understanding with the government building directly on the National AI Plan, committing to align any future Australian operations with the government’s published expectations. Officials described it as the first arrangement signed under the Plan, a marker the government has used to demonstrate that its sovereignty language is producing concrete commercial commitments rather than sitting only in a policy document.
Defence, Trilateral Ties, and the Wider Security Picture
AI’s role in Australia’s national security architecture has expanded in parallel with its civilian strategy. The 2026 Defence Innovation, Science and Technology Strategy, launched in August, names artificial intelligence as one of six focal technology areas, alongside quantum computing, autonomous systems, long-range fires, high-energy lasers, and undersea warfare, with platforms such as the Ghost Bat uncrewed aircraft and Ghost Shark autonomous undersea vehicle cited as concrete examples of where Defence expects AI-enabled systems to matter most. Deputy Prime Minister Richard Marles has framed the urgency in blunt terms, arguing that nations able to innovate faster and adopt new technologies more effectively will hold the decision advantage in future conflicts. That defence-side AI push runs alongside, and is reinforced by, Australia’s Trilateral Security Dialogue cooperation with the United States and Japan, which includes a dedicated AI Experts Group examining how the three countries can align AI capability development, address overlap and specialisation gaps, and strengthen resilience across shared AI value chains, work that sits adjacent to, but distinct from, the AUKUS submarine partnership’s own technology-sharing provisions.
What “Decoded” Looks Like From Here
Read together, Australia’s AI policy stack now runs on three tracks moving at different speeds. The National AI Plan sets broad economic and safety direction without yet constituting binding, cross-economy AI legislation, leaning instead on the AI Safety Institute, published infrastructure expectations, and voluntary industry commitments like the Anthropic agreement to do the work a statute might otherwise do. The APS AI Plan moves faster and more concretely, because government can mandate its own tools and training in a way it cannot yet mandate for the private sector. And the looming data centre and AI training standards, still working through National Cabinet as of mid-2026, represent the point where the plan’s sovereignty rhetoric will either convert into the promised “first in the world” legislated framework or reveal how much harder consistent national standards are to legislate across Australia’s federated state system than they are to announce. Nine months after the National AI Plan’s release, Australia has clearly moved past the ethics-framework and discussion-paper phase that defined its first decade of AI policy. What it has not yet fully answered is whether proportionate, follower-paced regulation can keep up with an AI infrastructure build-out that, on the government’s own projections, is set to roughly tenfold in the next decade.
Key References and Further Reading
- Australia’s AI Action Plan: Discussion Paper — Department of Industry, Science and Resources
- Australia’s Artificial Intelligence Plan Explained — Security & Defence PLuS
- AI in Australia’s Interests — Prime Minister of Australia
- Australia’s National AI Plan 2025: Key Takeaways — Clifford Chance
- AI Plan for the Australian Public Service 2025: At a Glance — Digital Transformation Agency
- A New Era for AI Governance in Australia: What the National AI Plan Means for Industry — Bird & Bird
- The Australian Public Service AI Plan 2025: A Legal and Commercial Roadmap — Norton Rose Fulbright
- Artificial Intelligence a Key Area of Australia’s 2026 Defence Innovation, Science and Technology Strategy — Cyber Daily
