Copyright, AI and the shifting onus on Australian creators
While there is little disputing the benefits that artificial intelligence (AI) brings to everyday life, its capabilities are advancing at a pace that can feel exhilarating and unsettling in equal measure. Case in point: the Australian public has only recently been alerted to the fact that OpenAI agents were breaching government websites seemingly undetected for many months. [1]
Yet alongside the promise of increased productivity, one question has quickly moved to the centre of the public debate: who has ownership and rights over the creative work that AI is built on?
For Australian creators and creatives, including artists, writers, musicians and journalists, as well as for the businesses developing and deploying AI, the answer matters enormously, and, right now, it is far from clear.
Key takeaways
- Copyright law as it currently stands was not built for the world of AI, with reform needed to ensure creators and artists are fairly compensated if their work is being used to train AI.
- Reforms should place the onus on those who wish to use creative work to seek the necessary permissions, rather than on those who created it.
- A recent landmark copyright deal was entered into by the Commonwealth Bank of Australia, by which it signed a business licence with a copyright agency, paving the way for companies to try and balance innovation and the fair treatment of rights holders in a mutually beneficial way.
A law that wasn’t built for this
Copyright is a form of intellectual property that is founded on someone’s creative skill and labour, and that protects original expression, be it a story, software, music, artwork – the list goes on.
The Copyright Act 1968 (Cth) is the sole source of copyright protection in Australia, [2] giving owners exclusive rights to reproduce, publish, communicate and adapt their work, [3] subject to certain fair-dealing exceptions. [4] Copyright generally arises from the time original work is created, and does not require registration. It is the licensing of copyright material that primarily produces economic returns for artists and creatives. There is no requirement to place a copyright notice or mark on protected works, but it does help to notify users of the existence of copyright, and to identify the copyright owner if someone is seeking permission to use that work.
Fundamentally, protection requires originality and material form. [5] In IceTV Pty Ltd v Nine Network Australia Pty Ltd (2009) 239 CLR 458, the High Court held that a work must originate with its author and reflect independent intellectual effort. In Telstra Corporation Ltd v Phone Directories Company Pty Ltd (2010) 194 FCR 142, the Full Federal Court found no copyright subsisted in largely automated directories, with Justice Yates observing that 'authorship' denotes human authorship. [6]
In other words, copyright exists to safeguard the expressions of human creativity. That fundamental principle is now under strain in the age of AI.
Training an AI model typically involves digitally copying vast quantities of text, images and audio from the world wide web, much of it protected by copyright. Under Australian law, that copying can amount to reproducing the work. [7] Whether any existing exception covers it is unsettled.
Content creators also have no practical way of knowing whether their work is or has been included in a training dataset. The scale of the problem was laid bare overseas when Anthropic agreed to pay US$1.5 billion to settle a claim over 482,000 books allegedly sourced from pirated online libraries without permission of, or payment to, copyright holders.[8]
The concern is shared closer to home. A 2025 survey by the National Association for the Visual Arts (NAVA) [9] of more than 890 visual artists found that, as at that time, over 80 per cent believed that AI posed risks to their income, practice and moral rights, and 73 per cent supported a compensation scheme when their work is used to train AI. Artists reported their work having already been scraped without their consent, and most respondents reported major barriers to identifying whether their work had been used at all, citing a lack of transparency, legal complexity and power imbalances. Notably, these artists did not reject the technology outright: almost half of those surveyed were already using generative AI in their creative practice, albeit largely for writing, research and administrative tasks.
Confusion around copyright ownership of AI-generated works has added another layer of vulnerability, with many artists expressing frustration over copyright protections relating to authorship and attribution, especially when AI had already been trained on human-made content without the permission of the creator.
The message from this survey is clear: artists seek the use of AI to be lawful, transparent and fair, underpinned by meaningful consent and clear avenues for attribution and compensation. [10] Clarifying issues around authorship and copyright ownership is key to ensuring accountability, attribution, and remuneration for creatives.
In the face of mounting pressure from creators, calls for legal certainty from the industry and amid the growing prevalence of AI, the Government has now stepped in to address this uncertainty.
Reforms on the table
In a speech at the University of Sydney on 15 July 2026, Prime Minister Anthony Albanese stepped squarely into the AI copyright debate. [11] He announced a dedicated Office of AI and promised that Australian writers, musicians, artists and journalists would keep ownership and control of their work, including control over its price and value. Our laws, he said, would spell that out 'plain as day'. No company, he insisted, should be able to use Australian creative work to build or train AI without the creator's say-so.
Those words are now being put to the test. The Australian Financial Review recently reported on proposals being considered by the Attorney-General's Department, which suggest how that promise might translate into law. [12]
The centrepiece is a 'Mandatory Data Access Arrangements' quota. Under this model, an AI company would first strike licensing deals with a minimum number of rights holders. Once it reached that threshold, the law would allow it to train on other unprotected material available online, without negotiating further deals or making further payments. Creators whose work falls outside those deals would receive nothing.
For creatives, the default position is stark. Their work would be available for AI training, without granting permission or recouping appropriate compensation, unless they take positive steps to opt out. The practical tools for doing so are blunt: locking work behind a paywall also locks out the very audience a creator is trying to reach.
And even for creators who do opt out, the proposal leaves one question unanswered: what about the work that has already been taken and used to train AI historically, for which there is no record?
The horse has bolted
There is a further uncomfortable reality. An opt-out regime is, by its nature, forward-looking. It does nothing for work that has already been ingested by AI models.
For many creators, the horse has already bolted: their novels, their songs, their photographs are already being utilised in training datasets, and no amount of opting out will extract them.
Where to from here?
None of this is to say that AI should be resisted. The companies developing AI are responding to a market that expects AI to rapidly grow, improve and learn. But the law’s task is to set the terms on which emerging and pervasive technologies interact with existing rights. On that score, Australia’s copyright framework has not kept pace.
Encouragingly, the market is already showing that a permission-based approach is workable. Just recently, the Commonwealth Bank of Australia became the first major corporate to sign a landmark copyright deal. [13] Specifically, CommBank signed a business licence with the not-for-profit Copyright Agency, allowing its more than 50,000 employees permission to use eligible Australian newspaper text and other covered content in prompts for workplace AI tools. In principle, licence fees paid under the agreement will flow back to creators and publishers, thereby investing in the creation of home-grown content. CommBank CEO Matt Comyn described the deal as a practical step showing that innovation and the fair treatment of rights holders can be mutually beneficial.
This agreement stands in contrast to the more flexible copyright laws the Government is being encouraged to consider by some AI juggernauts, including proposed opt-out models. CommBank’s deal may encourage other businesses to consider similar licensing arrangements as part of a proactive and responsible approach to AI.
Plainly reform is urgently needed. It should place the onus on those who wish to use creative work to seek the necessary permissions, rather than on those who created it to guard against its use.
Strong market forces are driving demand for AI to be trained, and to ‘learn’, at a rapid pace. As the CommBank deal shows, companies are trying both to advance in real time and to respond to that demand. Ideally, a balance will be struck between fostering innovation and protecting the creators of the very work on which that training depends.
In the meantime, creators need not wait for the law to catch up. Clearly identifying authorship is one practical step. A copyright notice and a statement that the work may not be licensed or used to train AI without permission makes it plain who owns the work and whose permission to seek before using it. It is no substitute for proactive reform, but it may help creators stand on firmer ground while reform takes shape.
After all, if the law cannot safeguard the human touch, what can? In an ever-changing technological world, its maintenance is more important than ever.
[1] Phillip Coorey, Nicola Smith and Paul Smith, ‘Rogue OpenAI Agent Breach Shows Australia Exposed on Cyber Defence’, The Australian Financial Review (online, 24 September 2026).
[2] Copyright Act 1968 (Cth), s 8 (The Act).
[3] Ibid s 31(1).
[4] Ibid ss 40-42.
[5] IceTV Pty Ltd v Nine Network Australia Pty Ltd (2009) 239 CLR 458, [33], [102].
[6] Telstra Corporation Ltd v Phone Directories Company Pty Ltd (2010) 194 FCR 142, [137].
[7]The Act ss 21(1A), 31(1)(a)(i).
[8] Bartz v Anthropic PBC (ND Cal, No 4:24-cv-05417, Martínez-Olguín J, 20 July 2026).
[9] Penelope Benton, ‘Artists Call for Consent and Compensation in AI Training’, National Association for the Visual Arts (Web Page, 2 September 2025).
[10] Ibid.
[11] The Hon Anthony Albanese MP, ‘AI in Australia’s interests’ (Speech, The University of Sydney, 15 July 2026).
[12] Paul Smith and Nicola Smith, 'Copyright Proposal Lets AI Giants Train for Free if Quotas Are Met', The Australian Financial Review (online, 15 September 2026).
[13] Danielle Long, ‘Commonwealth Bank Reaches a Landmark AI Deal with Copyright Agency to Pay for Content’, The Australian (online, 4 October 2026).
This article was prepared with the assistance of Nikki Young, Law Graduate.
Contact
