Key takeaways
- Australia is moving beyond regulating artificial intelligence (AI) at the point of deployment and use. The Australian Government is now developing additional national rules initially focussed on the infrastructure and inputs on which large-scale AI depends. The Government describes the proposed framework for large data centres and AI training as world leading, although its scope and legal architecture remain unsettled.
- The Australian Prime Minister, The Hon Anthony Albanese MP, has established a new Office of AI within the Department of the Prime Minister and Cabinet (PM&C), has proposed mandatory national AI standards, and has identified large data centres and Australian creative works as the first areas for legislation.
- The Prime Minister has subsequently indicated that the proposed data centre requirements would operate as conditions of approval for new proposals and are directed potentially at “very large” facilities. The precise coverage, legal form and enforcement arrangements have not yet been announced.
- On copyright, and in contrast to a number of other major economies, the Government has confirmed that it is not considering a broad text-and-data-mining exception and has committed to rights-holder control and payment. Licensing, offshore training, traceability and territorial reach remain to be addressed.
- Australia is pursuing a narrower approach than the European Union’s AI Act and a different approach from the fragmented United States framework. In effect, Australia is starting in those areas where it has practical regulatory leverage. The Government will seek to attract private investment in AI capability while imposing Australian requirements on the land, energy, water and protected works on which that investment depends. However, most deployed AI remains subject to existing Australian law.
The Prime Minister's speech
Australian AI policy is moving upstream towards infrastructure and training inputs. Until now, private-sector AI in Australia has been regulated principally through existing technology-neutral and sectoral laws, including privacy, consumer, competition, and employment laws.
The Prime Minister's speech (“PM’s Speech”) on 15 July 2026, titled "AI in Australia's interests", begins to place national conditions on the infrastructure and training inputs that make large-scale AI possible. The Government has established an Office of AI within PM&C, will seek National Cabinet agreement in August 2026 on mandatory national AI standards, and aims to bring legislation to Parliament in early 2027.
The first defined subjects are large data centres and the use of Australian creative works for model training. The broader coverage, legal architecture and enforcement arrangements remain to be developed. The Prime Minister’s subsequent interviews provide additional detail concerning the intended operation of the proposed framework.
(a) Office of AI
The most immediate change contemplated by the PM’s Speech is where responsibility now sits. The Office of AI has been established within PM&C to coordinate the design of the national standards, accelerate their implementation across the levels of government, and bring together work presently spread across industry, copyright, employment, education, energy, online safety, defence and national security.
Moving AI coordination into PM&C elevates AI from a portfolio technology issue to a whole-of-government strategic priority. The Prime Minister has clarified that the Office of AI will coordinate this work, while substantive government decisions will remain with Cabinet. The published materials do not identify licensing, investigation, compulsory audit or penalty powers for the Office, so it should presently be understood as a central policy and coordination body, rather than having an enforcement role. Whether enforcement functions are later created, and where they would sit, remains an open question.
(b) Mandatory national standards
The Government has committed to clear, consistent and mandatory Australian Standards for AI, to be developed through consideration at the National Cabinet forum and consultation with industry and Australia's trading partners. The Government is aiming to bring legislation to Parliament in early 2027.
In this context, the Government is using "standards" to describe requirements intended to be mandatory, rather than merely voluntary technical specifications. Beyond that, their eventual legal form and architecture remain unresolved. No exposure draft legislation has been published. The coverage, enforcement arrangements and penalty structure remain unsettled. No regulator has been identified as the principal enforcement body. Early comments suggest that access to computing capacity for Australian AI firms is one of the matters that the Government expects to approach through the standards.
The speech does not set out an Australian counterpart to the European Union's AI Act. It identifies no general mandatory private-sector framework for high-risk systems, model evaluation or AI-specific incident reporting. Rather, most uses of AI will, for now, remain subject to existing law, while new conditions begin to form around the infrastructure and inputs on which the AI economy depends.
(c) Proposed obligations for new data centres
The most concrete proposals concern large data centres. The Government proposes to convert central elements of its March 2026 expectations into legal obligations. The speech refers to the "next generation of large-scale data centres", without defining the class.
As noted in our previous article, the March 2026 Expectations of Data Centres and AI Infrastructure Developers, published by the Department of Industry, Science and Resources, apply more broadly to new or expanded co-location sites, hyperscale operations and large-scale AI compute centres, while excluding small-scale edge and on-site enterprise facilities. The legislation giving effect to the PM’s Speech will need to reconcile those formulations and define the treatment of projects that are approved, financed, contracted or being developed in stages.
In the PM’s Speech, covered data centre facilities would be required to underwrite their own new power supply, pay their full share of grid-connection costs so that energy bills are not affected, and put at least as much energy into the grid as they take out, supporting new renewable generation and firming capacity. The accompanying media release also describes an obligation to reduce power use when needed to strengthen the grid, together with requirements to be as water efficient as possible and to fund additional water infrastructure attributable to a project. The framework would also govern where covered facilities may be built, with input from local communities.
The Prime Minister wants data centres to become "net-generators, not net-users" and has warned that governments cannot revisit the issue after companies "have built whatever they want, wherever they want" and then attempt to re-open negotiations. The proposed obligations are also intended to create a stronger national framework for a sector currently regulated through state-based planning systems, electricity market rules, network connection processes and local approvals.
The Prime Minister has also said that the proposed obligations would operate as conditions of approval for new data centre proposals and should apply consistently across Australia, with the immediate focus on potentially “very large” facilities. This suggests that implementation may rely materially on state and territory approval regimes, although the relationship between those regimes, local planning processes and the proposed Commonwealth legislation remains unresolved. This raises questions about which authority will monitor ongoing energy and water commitments after construction and how nationally consistent standards will be maintained across different approval regimes.
(d) Copyright and AI training
The Prime Minister's comments on copyright are unusually categorical. Australian writers, musicians, artists and journalists must retain ownership and control of their work, including control over its price and value, and no company should use Australian books, music, art or news to build or train AI without the artist's control. Legislation is promised.
The Government had already confirmed, including in its 1 April 2026 Response to the Senate Select Committee on Adopting Artificial Intelligence, that the Government is not considering a text-and-data-mining exception. The Prime Minister has now gone further, stating that rights-holders should retain control and be paid if they agree to their work being used. The Prime Minister’s comments point towards consent and payment, potentially implemented through individual or collective licensing, rather than uncompensated access to protected material.
The mechanism remains open, and the Attorney-General’s continuing consultation through the Copyright and Artificial Intelligence Reference Group will inform its design. The PM’s Speech and subsequent interviews do not resolve whether licences would be negotiated individually or collectively, whether rights-holders could opt out of any collective arrangement, how offshore training would be reached, whether already trained models would be affected, or what disclosure and record-keeping duties would support enforcement.
Australia’s position differs from jurisdictions that provide statutory text-and-data-mining exceptions or apply broader fair use or fair dealing doctrines. The precise comparison depends on the jurisdiction and the purpose, source and circumstances of the copying. See our earlier article.
Two design questions will remain central. The legislation will need a territorial rule connecting the work, the rights-holder, the act of training or the supply of the resulting model to Australia. Effective enforcement will also depend heavily on traceability, because without records of training sources, rights-holders may be unable to identify or prove an infringing use, particularly where training occurred offshore.
Australia's strategy for the regulation of AI
Australia is unlikely to determine the rules under which the world's leading frontier models are built. A domestic model-specific regime might impose substantial costs on Australian organisations while exercising limited influence over models designed and trained offshore.
Australia has materially greater influence over the domestic conditions on which advanced AI depends, including such matters as land and planning, electricity and grid access, water, data centre approvals, copyright, foreign investment screening, government procurement, and communications infrastructure. The significance of the Prime Minister's speech lies in the Government’s decision to begin at two of those points: large-scale infrastructure and Australian creative works. The Prime Minister put the ambition directly: "We want Australia to have more of a stake in where AI is made and how it is made."
The emerging approach is conditions-based. The Government will seek to attract private investment in AI capability while imposing Australian requirements on the land, energy, water and protected works on which that investment depends. The interviews also suggest that sovereign capability will be pursued principally through private investment subject to Australian standards and approval conditions. Minister Ayres has since indicated that access to computing capacity for Australian AI firms is one of the matters the Government expects to consider through the mandatory standards. No access entitlement, allocation mechanism or pricing framework has yet been announced, and questions remain about affordable access for researchers, smaller business and public institutions.
Australian policy reached this point through four broad stages:
1. Voluntary principles in 2019
The Australian AI Ethics Principles provided guidance on fairness, transparency, contestability and accountability without creating enforceable obligations.
2. Analysis of regulatory gaps between 2023 and 2024
This produced the Voluntary AI Safety Standard, a proposals paper on mandatory guardrails for high-risk AI and a Senate committee recommendation for dedicated whole-of-economy legislation.
3. Consolidation around existing law
The National AI Plan of December 2025 placed existing Australian laws at the foundation of AI safety, supported by voluntary practice and an Australian AI Safety Institute. The Government did not proceed at that time with proposed mandatory AI guardrails. The Commonwealth adopted more prescriptive requirements for its own agencies: the Policy for the Responsible Use of AI in Government imposes mandatory oversight and accountability requirements on covered agencies, including impact assessments for covered use cases.
4. A focus on data centres and infrastructure
In March 2026, the Government published non-statutory expectations for data centres and AI infrastructure developers covering energy, water, Australian jobs and the national interest.
The PM’s Speech builds on these earlier initiatives while pursuing several objectives at once: investment attraction and faster, nationally consistent approvals; infrastructure that pays its own way in energy and water; protection of creators and news media; domestic skills and capability; national security and reduced dependence on concentrated foreign suppliers; and community legitimacy for the build-out.
International comparison - European Union
The European Union (EU) provides the clearest global example of comprehensive legislation that regulates AI. The AI Act prohibits certain practices, imposes transparency duties, regulates high-risk systems through risk management, documentation and conformity assessment, and imposes direct obligations on general-purpose AI models (with additional obligations for models presenting systemic risk).
Prohibitions under the AI Act began to apply in February 2025 and general-purpose model rules began to apply in August 2025. In June 2026, the European Parliament and the Council approved amendments moving the principal high-risk obligations to 2 December 2027 for stand-alone systems and to 2 August 2028 for systems embedded in regulated products.
The deferral reflected delays in the standards, guidance, authority designations and implementation capacity needed to administer the regime. However, Europe is also extending regulation into infrastructure. The proposed Cloud and AI Development Act would expand energy-efficient data-centre capacity and introduce an EU-wide framework for assessing cloud and AI sovereignty. The EU approach therefore increasingly reaches across models, uses and infrastructure.
International comparison - United States
The United States is often described as the more lightly regulated alternative to the EU, but its framework is fragmented. Deployed AI remains subject to federal consumer-protection, credit, employment, civil rights, intellectual property and sectoral laws. The National Institute of Standards and Technology (NIST) provides a voluntary risk-management framework and related guidelines. Litigation continues to shape the treatment of training data under fair use. States have adopted materially different approaches to automated decisions, discrimination, transparency and frontier-model safety.
Federal executive policy favours acceleration. The AI Action Plan of July 2025 and associated measures promote infrastructure build-out, energy supply, exports of the American technology stack and national-security capability. A June 2026 executive order established a voluntary framework for controlled early government access to, and cybersecurity evaluation of, covered frontier models, while expressly disclaiming mandatory licensing or pre-clearance. Executive Order 14365 of December 2025 directed federal challenges to state AI laws regarded as obstructive and called for pre-emptive national legislation.
The proposed Great American Artificial Intelligence Act, released as a discussion draft in June 2026 to solicit feedback before formal introduction, seeks to go further. It proposes federal obligations on large frontier developers, including safety frameworks, incident reporting and recurring third-party audits. The proposal would shift the regulation of model development principally to the federal level for three years, while preserving substantial state authority over post-deployment uses and generally applicable law. The current federal debate concerns whether model development should be subject to nationally uniform rules, while states retain authority over deployment, infrastructure and generally applicable laws.
What the speech means for Australian business
Further to our recent article discussing regulatory and policy developments, we highlight four key implications for Australian business:
Existing law continues to apply
From 10 December 2026, an Australian Privacy Principles (APP) entity using personal information in qualifying automated-decision-making arrangements must include specified information about those arrangements in its privacy policy. Organisations should identify the relevant systems and decisions before that date.
AI contracts require closer attention
Procurement terms should address the use of prompts, confidential information and customer data, further model training, training data provenance, copyright assurances, model changes, audit rights, indemnities, data portability, and termination assistance.
Infrastructure projects should be retested
Data-centre developers, investors and energy counterparties should determine whether a project is likely to constitute a new proposal and assess its structure against the proposed energy, grid, water and location conditions. Major compute users should consider the possible effects on capacity, pricing and continuity of supply.
Workplace AI deployment requires an existing-law review
No new statutory consultation requirement was announced, and workplace AI is being considered through a separate process led by the Employment Minister. Employers should assess such matters as existing consultation obligations under applicable awards and enterprise agreements, discrimination and privacy risks, and psychosocial work-health-and-safety duties before implementing consequential change.
What to watch next
Five key issues will determine the practical effect of the announcement:
1. Coverage and transition
The legislation will need to address concepts such as "new proposal", "large-scale" and "very large", and determine how expansions, staged projects and developments already in the approval pipeline will be treated. The energy obligations will also require measurement rules, including whether equivalent supply must be new, firm, local or available at matching times; how renewable generation is attributed to a facility; over what period compliance is assessed; and how curtailment is counted.
2. Legal architecture
The Prime Minister has indicated that the data-centre standards will operate as conditions of approval. The legislation must identify which governments and regulators will impose, monitor and enforce those conditions.
3. Copyright implementation
The Government will need to translate rights-holder control and payment into workable licensing, transparency and enforcement arrangements, including for offshore training and models developed from mixed international datasets.
4. Deployed AI and the workforce
The speech did not announce a general high-risk AI regime. Consumer-safety priorities are expected separately, while workplace AI is being considered through a process led by the Employment Minister involving business, unions, and other stakeholders. The relationship between those processes and the proposed national standards remains unclear.
5. Sovereign capability and economic return
The Government expects investment to be overwhelmingly private and has indicated that access to compute for Australian AI firms will be considered through broader standards. Questions remain about the form of that access, public procurement, Australian participation in supply chains, skills development, and the community benefits expected from large projects.
The PM’s Speech marks a material change in Australian AI policy, but the final framework remains to be designed. Its immediate significance lies in the Government’s decision to begin with data centre infrastructure and copyright, being two areas in which Australia has practical regulatory leverage. The legislation will determine whether those matters can be converted into nationally consistent and enforceable rules without deterring the investment and capability that the Government is seeking to attract.