Comment: Every day millions of people use Claude, ChatGPT, Gemini, DALL-E, Midjourney, Suno and many other AI models to create images, text, songs, code, and more, simply by prompting them. If you wanted to, you could prompt Suno to create an upbeat 80s pop song about riding your bike through Auckland’s CBD.
These AI models don’t magically create works from thin air. They are “trained” on vast amounts of songs, images, videos, text, code, and other sources of information, much of which is online and protected by copyright. Should AI companies pay the authors and other creators of the works they are training their models on? Debate about this has resulted in court cases around the world.
In New Zealand, Cabinet has asked the minister of commerce and consumer affairs to report back by March 31 2027 on a possible copyright framework for generative AI in New Zealand. If we want creators to receive some recompense for their work, I’d argue that New Zealand would be best to not follow the United States, which has been the site of the global battle of creators versus AI companies, and where there is a chance much of the wholesale copying of books and other works won’t infringe copyright. I’d suggest we look instead to Australia and the UK.
Bartz v Anthropic PBC was one of the first major US court cases to directly address whether AI companies can legally use copyrighted books to train large language models such as Claude. The plaintiffs were authors who alleged Anthropic had engaged in copyright infringement as it had copied their books without permission as part of building and training its AI systems. More than 400,000 authors and 15,000 publishers were represented.
Anthropic admitted scanning millions of hard copy books and downloading more than seven million books from pirated sites. In Bartz v Anthropic PBC Judge William Alsup quoted Anthropic CEO Dario Amodei as preferring to avoid the “legal/practice/business slog” of licensing or purchasing content through conventional channels.
For using pirated books Anthropic settled and agreed to pay the authors and publishers US1.5 billion. That might sound like a lot, but when dished out, each author received only $5000.
Anthropic infringed copyright, it paid for it. Open and shut case, right? If only. In law things are rarely so straightforward.
A concerning aspect of the Bartz v Anthropic PBC, certainly for authors and other creators, is that the court agreed with Anthropic’s argument that using hard copy books it had legally purchased and scanned to train Claude, amounted to “fair use” under US copyright law.
The reasoning was like the one that allowed Google Books to digitise millions of books without infringing copyright: the works were being used for a new purpose – creating a searchable index allowing users to locate words, terms and historical facts across millions of books.
Both cases involved strong legal arguments; however, as critics of the Bartz v Anthropic PBC would argue (myself included) it didn’t account for the potential market for or value of the copyrighted work. If AI companies can freely copy copyrighted works under the banner of fair use, authors and publishers lose the opportunity to license and therefore be paid for material that helps train these systems. Is that “fair use”?
New Zealand does not recognise fair use, instead we have fair dealing, which is considerably narrower than fair use. Anthropic’s actions would be extremely unlikely to amount to fair dealing in New Zealand.
Internationally AI companies are trying to pressure governments to change copyright law in their favour by creating text and data mining (TDM) exceptions, so they can use works on the internet, which are not behind paywalls, and physical items such as books to train their AI models without paying licence fees.
In response to the lobbying (from AI companies) the Australian and UK governments considered enacting TDM exceptions; however, strong responses from creatives saw both governments back down. In her press release, Australian Attorney-General Michelle Rowland said: “Artificial Intelligence presents significant opportunities for Australia and our economy, however it’s important that Australian creatives benefit from these opportunities too.
“Australian creatives are not only world class, but they are also the lifeblood of Australian culture, and we must ensure the right legal protections are in place.”
What about New Zealand? In July 2025, the Ministry of Business, Innovation & Employment released its “Responsible AI Guidance for Businesses” where it acknowledged that “fairly attributing and compensating creators and authors of copyright works can support continued creation, sharing, and availability of new works to support ongoing training and refinement of AI models and systems”.
If you agree with the ministry (and the Australian and UK governments) that authors’ and other creators’ work should not be copied and used without fair compensation, now might be the time to raise your voice, by emailing our Minister of Commerce and Consumer Affairs.
Professor Alex Sims will discuss the AI ‘training’ and copyright debate at Raising the Bar on Tuesday August 25
.png?w=600)
