Final week Disney, Common Footage and Warner Bros Discovery collectively sued MiniMax, a Chinese language synthetic intelligence (AI) firm, over alleged copyright infringement.
The three Hollywood media giants allege MiniMax (which operates Hailuo AI and is reportedly valued at US$4 billion) engaged in mass copyright infringement of characters akin to Darth Vader and Mickey Mouse by scraping huge quantities of copyrighted information to coach their fashions with out permission or cost.
This lawsuit is the newest in a rising record of copyright infringement circumstances involving AI. These circumstances embrace authors, publishers, newspapers, music labels and impartial musicians around the globe.
Disney, Common Footage and Warner Bros Discovery have the assets to litigate onerous and presumably form future precedent. They’re in search of damages and an injunction towards the continued use of their materials.
Circumstances like this one counsel the widespread strategy of “scraping first” and coping with penalties later could also be unsustainable. Different strategies for ethically, morally and legally acquiring information are urgently wanted.
One technique some persons are beginning to discover is licensed use. So what precisely does that imply – and is it actually an answer to the rising copyright issues AI presents?
What’s licensing?
Licensing is a authorized mechanism which permits using inventive works underneath agreed phrases, typically for a charge. It often includes two key gamers: the copyright proprietor (for instance, a film studio) and the consumer of the inventive work (for instance, an AI firm).
Typically, a non-exclusive licence is the place, in return for a charge, the copyright proprietor offers the consumer permission to train sure rights however retains possession of the work.
Within the context of generative AI use, granting a non-exclusive license may end in AI corporations gaining permission to be used and paying a charge. They may use the copyright proprietor’s materials for coaching functions, reasonably than merely scraping with out consent.
There are a number of licensing fashions, that are already being utilized in some AI contexts. These embrace voluntary, collective and statutory licensing fashions.
What are these fashions?
Voluntary licensing occurs when a copyright proprietor straight permits an AI firm to make use of their work, often for a cost. It could possibly work for big, high-value offers. For instance, the Related Press licensed their archive to OpenAI, the proprietor of ChatGPT.
Nonetheless, when there are millions of copyright homeowners concerned who every personal a smaller variety of works, this technique is gradual, cumbersome and costly.
One other drawback is that after a generative AI firm has made one copy of a piece underneath license, it’s unsure whether or not this copy could also be used for different duties. Additionally, making use of voluntary licensing to AI coaching is tough to scale, as a result of coaching requires huge datasets.
This makes particular person agreements with every copyright proprietor impractical. It may be advanced by way of figuring out who owns the rights, what needs to be cleared and the way a lot to pay. The licensing charge may be prohibitive to smaller AI companies, and particular person copyright homeowners could not obtain a lot income for the use.
Collective licensing permits copyright homeowners to have their rights managed by an organisation often known as a amassing society. The society negotiates with the consumer and distributes licensing charges to the copyright homeowners.
This mannequin is already generally used within the publishing and music industries. In idea, whether it is expanded to the AI trade, it may present AI corporations with entry to giant catalogues of information extra effectively.
There are already some examples. In April 2025, a collective license for generative AI use was introduced in the UK. Earlier this month, one other was introduced in Sweden.
Nonetheless, this mannequin raises questions on charge buildings, and the precise use itself. How would charges be calculated? How a lot can be paid? What constitutes “use” in AI coaching? It’s unsure whether or not copyright homeowners with smaller catalogues would profit as a lot as huge gamers.
A statutory (or obligatory) licensing scheme is another choice. It already exists in different contexts in Australia akin to training and authorities use. Below such a mannequin, the federal government may allow AI companies to make use of works for coaching with out requiring permission from every copyright proprietor.
A charge can be paid right into a central scheme at a predetermined fee. This strategy would guarantee AI corporations entry coaching information whereas making certain some remuneration to copyright homeowners. Nonetheless, it removes copyright homeowners’ capability to say no to the use.
A danger of domination
In apply, these licensing fashions sit on a spectrum with variations. Collectively, they symbolize some future methods the rights of creators could also be reconciled with AI corporations’ starvation for information.
Totally different types of licensing provide potential alternatives for copyright homeowners and AI corporations. It’s certainly not a silver bullet.
Voluntary agreements will be gradual, fragmented and never end in a lot income for copyright homeowners. Collective schemes elevate questions on equity and transparency. Statutory fashions danger under-valuing inventive work and rendering copyright homeowners powerless over using their work.
These challenges spotlight a a lot larger subject which is raised when copyright is taken into account in new technological contexts. That’s, methods to strike a stability between these concerned, whereas nonetheless selling equity and innovation.
If a cautious stability just isn’t struck, there’s a danger of domination from a handful of highly effective AI corporations and media giants.
Wellett Potter, Lecturer in Regulation, College of New England
This text is republished from The Dialog underneath a Artistic Commons license. Learn the unique article.
