AI versus artists: the Australian copyright debate

Words: Dr Jonathan Harlen
Published: 29 September 2025

Southern Cross University Law Lecturer and intellectual property specialist Dr Jonathan Harlen examines how the rapid rise of artificial intelligence has left Australian copyright law scrambling to keep up.

Proposed reforms, including a limited “fair use” exception for text and data mining, could put creators at risk of seeing their intellectual property absorbed into the next wave of AI systems.

Across the globe, disputes are unfolding between tech giants, publishers, and creative artists over whether copyrighted material can be used to train AI systems like ChatGPT and Midjourney. In Australia, the debate is now heating up, as policymakers consider whether to introduce a new “fair use” exception to copyright for large-scale text and data mining.

Multiple lawsuits are currently playing out in the United States, Europe and China pitting AI tech giants against publishers, media companies and creative artists.

Copyright holders are alleging that their original works have been ‘scraped’ (copied in bulk from websites, databases and digital platforms by automated programs) on an unprecedented scale, to be used as training materials for increasingly sophisticated AI chatbots and text-to-image generators such as ChatGPT, Claude and Midjourney.

In early September, one of the most high-profile defendants, Anthropic PBC, agreed to pay $1.5 billion to settle a class action by three book authors, on behalf of potentially millions of others. The settlement followed a ruling by the US District Court for the Northern District of California that the training of Anthropic’s chatbot Claude was not in itself illegal, however the pirating and storage of more than half a million original works for use in Claude’s training did indeed infringe the authors' copyright.

The case was on appeal to the US Ninth Circuit when it settled. Had the District Court ruling on the pirated works been upheld, Anthropic could have been found liable for damages totalling US $75 billion, more than enough to ruin the company. As it stands, the correctness of the District Court ruling remains uncertain, since it has not been confirmed by a higher court. Other key legal issues continue to remain unresolved, such as the status of works created by AI that are substantially similar to the original works they were trained on. The legal arguments surrounding the latest generation of generative AI tools are still a long way from being resolved.

In Australia the debate over these issues heated up considerably following the release of the Productivity Commission's Interim Report on Harnessing Data and Digital Technology in August this year. The Commission indicated in the report that it is considering recommending a limited “fair use” exception to copyright, to cover text and data mining (the so-called ‘TDM’ exception).

The Commission confirmed that this ‘TDM’ exception, if enacted, would cover not just the scraping of data to train AI chatbots and other large language models, but all kinds of other analytical techniques that use machine-read material to identify useful patterns and trends. The Commission has called for further submissions ahead of its final report to be released in December this year.

Historically, Australian copyright law has allowed only a small number of specific and limited “fair dealings” as exceptions to copyright. These include fair dealings for research and study, for reporting the news, for criticism and review, for satire and parody, and for legal advice.

Southern Cross University Law Lecturer and intellectual property specialist Dr Jonathan Harlen

Prominent Australian artists such as Virginia Trioli and Peter Garrett have condemned the proposal, as have heavyweight arts institutions such as APRA-AMCOS, the Australian Publishers Association, and Creative Australia. By contrast, Australian billionaire and co-founder of software company Atlassian, Scott Farquhar, has called on the Australian government to legislate for the exception, arguing that it would generate “billions of dollars” in foreign investment and give Australia a significant boost in the global AI arms race.

In early August one of Australia’s leading AI entrepreneurs, Dave Lemphers, proposed a solution which now seems remarkably timely, given the subsequent settlement of the Anthropic case. Lemphers is currently building what he claims is Australia's answer to ChatGPT, and he remains convinced that it is possible to train AI to a very high standard without infringing copyright, and without making radical changes to Australia’s existing copyright laws.

What is needed, he argues, is a government-mandated royalty-collection system to oversee a tailored licensing regime designed to reward Australian copyright owners whose data contributes to AI systems.

According to Lemphers, the position is not radical due to the royalty-based systems already in place for industries including music, publishing and IP licensing. He states that the key difference is the scale and importance of the downstream product.

Just weeks later, in September this year, an agreement by Anthropic was reached to divide a total of $1.5 billion in damages among an estimated half a million plaintiffs, sending a very strong signal to other big tech companies to move toward similar licensing regimes. Companies who do not agree to pay license fees to copyright holders now run the risk of incurring enormous liabilities, should they leave these disputes in the hands of the courts.

For so long on the back foot, artists, publishers, media companies and other creatives are at last beginning to wrestle some measure of control over the use of their original works by their AI competitors. This shift signals the beginning of a more structured relationship between human creativity and machine learning , and we can expect to see more licensing arrangements being announced in the not-too-distant future.

Banner artwork by Nidia Dias, Google DeepMind.