Right now in the United States, there are dozens of pending lawsuits involving copyright claims against artificial intelligence (AI) platforms. The judge in one case summed up what’s on the line when he said:
These products are expected to generate billions, even trillions, of dollars for the companies that are developing them. 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.
On each side, the stakes seem existential. Authors’ livelihoods are at risk. Copyright-based industries – publishing, music, film, photography, design, television, software, computer games – face obliteration, as generative AI platforms scrape, copy and analyse massive amounts of copyright-protected content.
They often do this without paying for it, generating substitutes for material that would otherwise be made by human creators. On the other side, some in the tech sector say copyright is holding up the development of AI models and products.
And the battle lines are getting closer to home. In August, the Australian Productivity Commission suggested in an interim report, Harnessing Data and Digital Technology, that Australia’s copyright law could add a “fair dealing” exception to cover text and data mining.
“Fair dealing” is a defence against copyright infringement. It applies to specific purposes, such as quotation for news reporting, criticism and reviews. (Australian law also includes parody and satire as fair dealing, which isn’t currently the case in New Zealand).
While it’s not obvious a court would agree with the commission’s idea, such a fair dealing provision could allow AI businesses to use copyright-protected material without paying a cent.
Understandably, the Australian creative sector quickly objected, and Arts Minister Tony Burke said there were no plans to weaken existing copyright law.
On the other hand, some believe gutting the rights of copyright owners is needed for national tech sectors to compete in the rapidly developing world of AI. A few countries, including Japan and Singapore, have amended their copyright laws to be more “AI friendly”, with the hope of attracting new AI business.
European laws also permit some forms of text and data mining. In the US, AI firms are trying to persuade courts that AI training doesn’t infringe copyright, but is a “fair use”.
An ethical approach
So far, the New Zealand government has not indicated it wants similar changes to copyright laws. A July 2025 paper from the Ministry of Business, Innovation and Employment (MBIE), Responsible AI Guidance for Businesses, said:
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.
MBIE also has guidance on how to “ethically source datasets, including copyright works”, and about “respecting te reo Māori (Māori language), Māori imagery, tikanga, and other mātauranga (knowledge) and Māori data”.
An ethical approach has a lot going for it. When a court finds using copyright-protected material without compensation to be “fair”, copyright owners can neither object nor get paid.
If fair dealing applied to AI models, copyright owners would basically become unwilling donors of AI firms’ seed capital. They wouldn’t even get a tax deduction!
The ethical approach is also market friendly because it works through licensing. In much the same way a shop or bar pays a fee to play background music, AI licences would help copyright owners earn an income. In turn, that income supports more creativity.
Building a licensing market
There is already a growing licensing market for text and data mining. Around the world, creative industries have been designing innovative licensing products for AI training models. Similar developments are under way in New Zealand.
Licensing offers hope that the economic benefits of AI technologies can be shared better. In New Zealand, it can help with appropriate use of Māori content in ways uncontrolled data scraping and copying don’t.
But getting new licensing markets for creative material up and running takes time, effort and investment, and this is especially true for content used by AI firms.
In the case of print material, for example, licences from authors and publishers would be needed. Next, different licences would be designed for different kinds of AI firms. The income earned by authors and publishers has to be proportionate to the use.
Accountability, monitoring and transparency systems will all need to be designed. None of this is cheap or easy, but it is happening. And having something to sell is the best incentive for investing in designing functioning markets.
But having nothing to sell – which is effectively what happens if AI use becomes fair dealing under copyright law – destroys the incentive to invest in market-based solutions to AI’s opportunities and challenges.
Graeme Austin does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.
This article was originally published on The Conversation. Read the original article.
