Copyright and AI training in Australia: Insights from US cases

As artificial intelligence (AI) becomes increasingly embedded in our everyday lives, its rapid development is fuelled by the relentless training of large language models (LLMs) on vast datasets.

Beneath the surface of this progress lies a growing controversy which is the unauthorised use of copyright works to train AI datasets. Recent US court rulings in favour of tech giants Meta¹ and Anthropic² have brought this industry practice into the global spotlight, sparking serious concerns for creators and businesses, particularly in Australia.

Case summary - Meta

On 25 June 2025, a federal judge in California granted Meta's application for a partial summary judgment in proceedings commenced by 13 authors who alleged that Meta had infringed their copyright by using their books without permission to train Meta's LLM, LLaMa.

Meta downloaded the books and other text from shadow libraries after failing to negotiate licensing agreements with authors and publishers.

Meta argued that its use of the books qualified as fair use, a defence to copyright infringement under US law, on the basis that the AI training was 'transformative' in nature. The plaintiffs contended that the use did not qualify as fair use and asserted that LLaMa was capable of reproducing small snippets of text from their books and that such use diminished their ability to licence their works for the purpose of training LLMs.

Judge Chhabria dismissed the plaintiffs' arguments as "clear losers", particularly after Meta had implemented 'mitigations' to prevent the LLM from memorising and reproducing copyright material. He also noted that the plaintiffs' expert could only extract a maximum of 50 words from their books through the LLM.  Judge Chhabria stated that the plaintiffs have no entitlement to monopolise the licensing market for their works.

In response to arguments that restricting copyright would "stop the development of this groundbreaking technology", Judge Chhabria added that "if using copyrighted works to train the models is as necessary as the companies say, they will figure out a way to compensate copyright holders for it".

The judge emphasised that "in many circumstances it will be illegal to copy copyright-protected works to train generative AI models without permission", making clear his decision was confined to the specific facts.

Case summary – Anthropic PBC

Just two days before the judgment in Meta's favour, a similar decision was handed down in support of Anthropic PBC, an AI software firm founded by former OpenAI employees. In a class action lawsuit, authors alleged that Anthropic used their books without authorisation to train their GenAI system, Claude, by initially downloading pirated copies and later purchasing and scanning the books.

Judge Alsup determined that the latter use of the lawfully purchased books qualified as fair use, comparing Claude to a human who learns from copyright material.  He stated that "the technology at issue was among the most transformative many of us will see in our lifetimes". The court will proceed to trial in relation to the earlier use of the pirated copies of the books.

Application under Australian law

While Australian courts have yet to hear any cases involving legal issues related to AI, it's likely that the outcome will be different under Australian copyright law as we do not have a fair use defence.

Instead, the fair dealing defences under the Copyright Act 1968 (Cth), only apply to specific purposes such as research or study, criticism or review, parody or satire and reporting of news. These exceptions are unlikely to apply in a commercial AI training context, especially since any use must be considered 'fair' to qualify.

In Australia, the copyright in a literary work is infringed when a substantial part of the work is reproduced in material form, which would include where a pirated book is downloaded from the internet, without the licence of the copyright owner. It is likely that Australian courts would consider the unauthorised digital reproduction of books for AI training to constitute copyright infringement.

Given Australian copyright law is more aligned with the UK, the decision in the proceedings commenced by Getty Images against Stability AI would have been of interest.  However, Getty recently decided during its closing submissions to withdraw its copyright infringement claims relating to the use of images as data inputs for the purposes of training and developing Stability AI, largely because those activities took place outside the UK.  This is likely to present an issue for Australian copyright owners whose works are used for AI training, given this conduct largely occurs offshore.  Getty has also commenced proceedings against Stability AI in the US.

What your business can do

Without a fair use defence, Australian law presents a tighter rope for AI developers and businesses who want to use copyright material to train and develop generative AI.

That said, the use of copyright works for AI training is permissible where a licence has been obtained from the copyright owner. For example, Australian tech giant, Canva has introduced the Creator Compensation Program which pays creators who consent to have their content used to train its AI models.  They also offer proactive consent for users to opt out of having their designs used to train Canva's AI models.

Initiatives such as this are likely to more appropriately balance the importance of technological development with the need to adequately compensate copyright owners for the use of their works.

If you would like any assistance on copyright and AI issues, please contact our Intellectual Property or Technology teams.

¹ Kadreyv. Meta Platforms, Inc., 3:23-cv-03417, (N.D. Cal., 2025)

² Bartz v. AnthropicPBC, 3:24-cv-05417, (N.D. Cal., 2025)

Authors

Marlia Saunders | Partner | +61 2 8248 5836 | msaunders@tglaw.com.au

Dianne Beer | Special Counsel | +61 2 8248 5816 | debeer@tglaw.com.au

Andrea Roque | Law Graduate

‍