Introduction

Australia has taken a significant step aimed at becoming a global AI investment hub, announcing the creation of a new Office of Artificial Intelligence within the Department of the Prime Minister and Cabinet.

While the announcement does not introduce a dedicated AI Act or new compliance obligations, it signals a major shift in how the Commonwealth intends to coordinate AI policy, attract investment and position Australia in the global race for AI infrastructure.

In a speech delivered at the University of Sydney on 15 July 2026, titled AI in Australia's Interests, Prime Minister Anthony Albanese announced the creation of the Office and that it would commence operation immediately.

The Office will coordinate the design of new Australian Standards for AI and bring together work currently spread across portfolios, spanning energy and data centres, intellectual property, productivity, education and labour rights.

The Government has framed the initiative as a world first: a single, national framework for AI, positioned to give international investors greater clarity, faster approvals and a streamlined process for verifying compliance.

The announcement matters as much for what it does not contain as for what it does.

Coordination, not new law

The Office of AI is a coordination mechanism rather than a source of new substantive obligations, and it is consistent with the trajectory set by the National AI Plan in December 2025: no dedicated AI Act, no mandatory guardrails for high-risk settings, and regulation through existing legal frameworks with targeted sector-specific interventions layered on top.

The obligations that matter for organisations deploying AI remain those already in force, including the Privacy Act (with automated decision-making transparency obligations commencing 10 December 2026), the Australian Consumer Law, directors' duties under the Corporations Act, and the Copyright Act.

The Office will work closely with the Industry and Innovation Minister and the Assistant Minister for Science, who in November 2025 released national principles for AI data centre approvals under which developers are expected to fund their own firmed renewable energy in exchange for fast-tracked approvals.

The copyright question remains open

Conspicuously, the announcement contains no change to copyright law. In October 2025, the Government rejected the Productivity Commission's proposal for a text and data mining exception to the Copyright Act 1968 (Cth), which would have permitted AI developers to train models on copyright works without permission or payment.

The position in Australia therefore remains that training an AI model on copyright material requires a licence. Australia has no general fair use doctrine, and the Attorney-General's Copyright and AI Reference Group is instead examining licensing frameworks (including a possible paid collective licensing model), the treatment of AI-generated outputs, and a low-cost small claims mechanism for copyright enforcement.

The territorial dimension of that position was illustrated by the UK High Court's November 2025 judgment in Getty Images v Stability AI [2025] EWHC 2863 (Ch). Getty abandoned its primary infringement claims over model training because there was no evidence that the training of Stable Diffusion had taken place in the United Kingdom, and its remaining secondary infringement claim failed because the model weights do not store or reproduce the works on which the model was trained.

The lesson is that copyright is territorial: the lawfulness of AI training is governed by the law of the place where the copying occurs. Training run on United States infrastructure is assessed against the US fair use doctrine, while the same training conducted in Australian data centres would be assessed against the Copyright Act 1968 (Cth), with no text and data mining exception and no general fair use defence.

Jurisdictional copyright settings are therefore a direct input into where AI companies locate their training compute and operations, which is precisely why copyright clarity and data centre investment have arrived in Canberra as a single, linked question.

The commercial stakes of that settled position were laid bare in the days before the announcement. Briefing notes released under freedom of information laws revealed that Anthropic has told the Treasurer that its proposed $21.6 billion investment in Australian AI infrastructure, including some 1.4 gigawatts of data centre capacity, is contingent on clarity of copyright settings and certainty over its liability to rights holders.

Treasury's advice pushed back on the suggestion that AI training is settled 'fair use', noting the volume of active copyright litigation in the United States, including Anthropic's own US$1.5 billion settlement with authors in 2025. The Prime Minister's response is that the right guardrails will attract investment rather than repel it.

The commercial picture – Will Australia profit from AI dollars?

The question lands at a significant moment for the national economy.

Iron ore has been Australia's largest export for two decades, but the official outlook now points firmly downward, with the Department of Industry's June 2026 forecasts showing iron ore export earnings falling from $117 billion in 2025-26 to around $77 billion in real terms by 2030-31 as Chinese steel production moves past its peak.

The search for what fills that gap has become a live economic policy question, and AI infrastructure is an obvious candidate: Australia's endowments map well onto what large-scale compute requires, including land, energy resources, political stability, alliance relationships and growing regional demand for data residency. On that view, the proposal is less a one-off transaction than an early test of whether Australia can convert those endowments into a durable new pillar of the economy as the resources engine slows.

Beneath the policy debate sits a harder commercial question, one crystallised in recent Australian Financial Review commentary: whether Australia captures durable economic value from hosting AI infrastructure, or simply supplies the land, energy and water while the productivity dividend accrues offshore.

The 1.4 gigawatts of capacity reportedly sought is roughly equivalent to the entire existing Australian data centre industry, requests for proposal have gone to five local operators with the mandate potentially split across several providers, and a development of that size is estimated to require USD12 to 15 billion in debt and equity financing.

Iron ore delivered national returns through a settled royalty and taxation architecture; no equivalent mechanism exists for hosting AI compute, so the legal and contractual settings on which the investment arrives become the primary instruments of value capture.

The commercial risk allocation deserves as much attention as the policy settings.

Concentration and sovereign risk are equally live, with United States export controls briefly interrupting Australian corporate access to a frontier-tier model in June 2026, a reminder that access to the technology remains subject to foreign regulatory action however much capital is deployed locally. The reported architecture of any eventual bargain, pairing large-scale infrastructure investment with an annual fund for creators, has already drawn political fire, and the Government has rejected suggestions that any such arrangement has been agreed.

What to watch

There will be plenty of activity in the AI space in Australia in the coming months and we'll be monitoring:

At Thomsons, our Technology and Digital Innovation team advises AI developers, deployers, rights holders and infrastructure investors across the full range of these issues. If you would like to discuss what the new framework means for your organisation, please contact us.

Read more on AI and the key issues we are tracking on our Artificial intellegence and legal design key topic page.