Artificial intelligence and copyright in Australia: from consultation to commitment

Rebecca Dunn
Gilbert + Tobin, Sydney
rdunn@gtlaw.com.au

Caitlin Meade
Gilbert + Tobin, Sydney
cmeade@gtlaw.com.au

Introduction

The question of how intellectual property should respond to artificial intelligence (AI) has divided legislatures and courts worldwide. The US is testing the boundaries of fair use through litigation, the European Union has legislated a text and data mining regime with an opt-out option and the UK continues to consider its approach through successive consultations. In Australia, the federal government has declined to introduce a text and data mining exception for the use of copyright protected works in the training of AI models and has announced the creation of a mandatory national AI framework that places the protection of rightsholders at its centre.

This article traces the developments that has brought Australia to this position: the structural features of the Copyright Act 1968 (Cth) that frame the debate, the Productivity Commission’s proposal for a text and data mining exception and its rejection, the government’s announcement that new Australian AI standards and a new Office of AI will be established and the doctrinal questions that remain unresolved.

The Copyright Act 1968 (Cth): a closed system of exceptions

The Australian Copyright Act confers on owners the exclusive right to reproduce a work in material form and to communicate it to the public. Australian copyright law contains no general fair use defence and provides only enumerated fair dealing exceptions for defined purposes, such as research or study, criticism or review, parody or satire and reporting news.[1] None of those purposes accommodates the commercial-scale reproduction of works involved in training a generative AI model. The consequence, for orthodox principles, is that training a model on copyright works within Australia without a licence will involve infringing reproduction, subject to proof that copying of a substantial part occurred during the training process.

However, the territorial reach of the Copyright Act complicates that picture. Australian copyright law governs acts conducted in the territory of Australia, and most large-scale model training to date has occurred offshore. The practical exposure of AI developers under Australian law has, therefore, centred on outputs and local deployment rather than the historical training that has occurred in other jurisdictions and which has generated litigation in the US, Germany and elsewhere. That territorial gap is precisely what has made the policy question so pointed: whether Australia should liberalise its exceptions to attract onshore AI development or hold its settings, maintain the current protection for rightsholders and enable the building of licensing infrastructure instead. The past year has answered that question, at least for now, in favour of the latter.

The Productivity Commission and the text and data mining debate

The liberalisation case was put most prominently by the Productivity Commission. In its interim report Harnessing Data and Digital Technology, released on 5 August 2025 as part of the government’s productivity growth agenda, the Commission estimated that AI could add in the order of AU$116bn to Australian gross domestic product in over a decade and identified the copyright regime as an area where government could deliver regulatory clarity. Among the policy options it advanced for consultation was the introduction of a text and data mining exception, framed as a new fair dealing purpose or a broader fair use style defence, which would have permitted the use of copyright material to train AI models in Australia.[2]

The Productivity Commission’s proposal provoked a forceful response from the creative sector, strengthened by revelations earlier in 2025 that the works of many Australian authors had been included in pirated datasets used to train large language models. Copyright collecting society APRA AMCOS (made up of the Australasian Performing Right Association (APRA) and the Australasian Mechanical Copyright Owners Society (AMCOS)) characterised the Productivity Commission’s recommendation as an attempt to ‘legitimise digital piracy under guise of productivity’.

However, in October 2025 the Attorney-General rejected the proposed exception outright, confirming that Australian copyright law would not be amended to permit free training on copyright works.[3]

The Commission’s final report, released on 19 December 2025, accepted the government’s position. It recommended against the immediate introduction of a text and data mining exception, preferring a three-year monitoring approach, and cautioned more generally against AI-specific regulation unless a gap analysis demonstrated that existing law was inadequate.[4]  Following this, the debate between rightsholders and AI developers, deployers and those in favour of incentivising AI investment has shifted to the design of licensing solutions.

The national AI framework: the July 2026 announcement

The most significant development came recently on 15 July 2026, when Australian Prime Minister Anthony Albanese delivered a keynote speech in Sydney titled ‘AI in Australia’s Interests’, setting out a new whole-of-government policy direction.[5] The framework has three principal elements. First, the government will introduce mandatory Australian AI standards, described as the first attempt internationally to bring economic, social, copyright and infrastructure issues within a single national framework. Secondly, a new Office of AI has been established within the Department of the Prime Minister and Cabinet to coordinate the design of the standards across the relevant portfolios. Thirdly, the framework carries specific commitments on copyright. The Prime Minister said that:

‘But let me make this crystal clear: not everything produced in Australia is up for grabs. Not at all. Australian writers, musicians, artists and journalists must retain ownership and control of their work. Our laws will spell that out, plain as day. An artist’s creative endeavour is their work and their property. No company should use Australian books, music, art or news to build or train AI without the artist’s control. That includes the artist’s control of the price and value of their work. Anything less, is theft.’

The Attorney-General’s consultation on copyright and AI continues through the Copyright and Artificial Intelligence Reference Group, the standing stakeholder mechanism established in December 2023, with three principal models under consideration: statutory licensing, collective licensing and voluntary arrangements. Bilateral licensing between individual rightsholders and AI developers is difficult to achieve at the scale that model training requires, and the Minister for Industry and Innovation has framed the government’s intervention as breaking a stalemate that market negotiation alone could not resolve. Some form of statutory or collectively administered scheme, therefore, appears the most likely outcome. The Prime Minister will seek the agreement of the states and territories to the mandatory standards before the National Cabinet in August 2026, and implementing legislation is expected to reach Parliament in early 2027.

For rightsholders, the framework represents an endorsement of consent-based use, and positions Australia at the protective end of the international spectrum, in contrast to the EU’s opt-out text and data mining regime and the fair use arguments being tested in US litigation. For AI developers, it signals that access to Australian works will come at a price set through licensing rather than through exceptions, although the detail of the scheme, including its treatment of offshore training and its interaction with the territorial limits of the Copyright Act, remains to be settled.

Authorisation, authorship and the unresolved doctrinal questions

There remain critical questions about the way that existing copyright laws in Australia apply to generative AI’s use and the possible ‘creation’ of works. In the absence of a text and data mining exception or the broad ‘fair use’ doctrine present in other jurisdictions, the training of generative AI tools is, prima facie, likely to involve infringement under the Copyright Act, however, to establish that an infringement has occurred factual questions about the precise operation of those tools and the precise nature of the reproduction will need to be answered. In addition, questions of secondary liability are likely to assume renewed importance. Australian law on the authorisation of infringement, shaped by the High Court’s decision in Roadshow Films Pty Ltd v iiNet Ltd, turns on the power a party has to prevent an infringement and the reasonableness of the steps it has taken,[6] and the application of that framework to providers of AI platforms whose tools can be used to generate infringing outputs is an open and consequential question.

On the protection of products generated by AI tools, the human authorship requirement articulated in IceTV Pty Ltd v Nine Network Australia Pty Ltd and Telstra Corporation Ltd v Phone Directories Co Pty Ltd points strongly to the conclusion that purely AI-generated outputs attract no copyright protection in Australia, although the degree of human contribution through prompting, selection and refinement that would suffice for authorship remains untested.[7]

Conclusion

Australia has now adopted a distinct course of action in regard to the global debate on AI and intellectual property. By rejecting a text and data mining exception and committing instead to a consent-based, licensing framework supported by mandatory national AI standards, the government has signalled that the interests of rightsholders will remain a central feature of Australia’s AI policy.

That policy choice, however, does not resolve the difficult legal questions that generative AI presents. Existing copyright principles concerning infringement, authorisation and human authorship will be tested in new and complex factual settings. As the government develops the detail of its proposed licensing framework and litigation emerges, Australia will be an important case study on the way established intellectual property principles can respond to generative AI.