AI Copyright Licensing: Market Solutions to GAI Development
Generative Artificial Intelligence (GAI) technologies are improving and providing more generative capabilities at a rapid pace. Their continued development will depend on the viability of burgeoning markets in which copyright owners license copyright works for GAI training. The existing licensing agreements between AI companies and rights holders exemplify how the two communities can work together and ensure that strong IP and strong AI co-exist. Just take a look at this list of different types of copyright holders who reached diverse licensing agreements with AI companies, demonstrating not only a thriving licensing ecosystem, but also a future where AI technologies and creativity flourish together.
Copyright Licensing Benefits Creativity and the Economy
Core copyright industries contributed $2 trillion to the United States’ GDP in 2023. The significant economic impact of American copyright industries flows from the exclusive rights granted under copyright law and the ability of rights holders to license those rights. Advancement in technology is not an excuse to undermine copyright owners’ rights. Indeed, while past advances in technology, e.g. streaming, raised many questions about the distribution of copyrighted works and licensing, copyright owners—supported by a judicial system that properly recognized their rights—and platforms worked together to answer those questions and develop the vibrant streaming environment we enjoy today.
Copyright licensing encourages creativity and innovation by rewarding copyright owners and providing AI companies with quality, curated training material. Indiscriminately scraping online sources risks exposure to illegal and/or low-quality works to train GAI. This results in GAI models generating outputs that infringe copyright and other intellectual property, violate privacy and right of publicity laws, spew propaganda or discriminatory language, generate misleading information, and otherwise result in harmful content. Licensing overwhelmingly reduces these risks by offering fact-checked, high-quality training material in the form of copyrighted works. The goal of gold standard GAI can be achieved by training GAI models with licensed copyrighted works.
Direct Voluntary AI Licensing Markets are Thriving
Licensing can satisfy demands for a variety of uses of a copyrighted work, and specific licensing terms can and are being agreed to for GAI training. These types of licensing agreements ensure the development of high functioning generative models without devaluing copyright law and the works it protects.
The diverse licensing options for rightsholders and AI companies that support ethical innovation continue to expand. The Association of American Publishers’ (AAP) amicus curiae in Kadrey v. Meta lists over 30 direct licensing agreements between textual works rights holders and AI companies—demonstrating how recent court holdings, which entirely dismissed the AI licensing market, would harm a $2.5 billion market projected to reach $30 billion within the decade.
News media publishers have been at the forefront of AI licensing. These are just a few of those agreements:
- Associated Press with OpenAI and Google
- Axel Springer with OpenAI and Microsoft
- Financial Times with OpenAI
- New York Times with Amazon
- Reuters with Meta
When courts wrongly reject the existence and significance of a clearly robust AI copyright licensing market through faulty fair use analyses, they threaten to undermine the rights of copyright holders and adversely impact the existing and future licenses between these copyright holders and AI companies.
Copyright owners of literary works are not the only participants of this thriving AI licensing market. GAI technologies are developing beyond the large language model context, leading other copyright owners of a variety of works of authorship to negotiate and enter licensing agreements. For example, music companies including Warner, Universal Music Group (UMG), and Sony are negotiating licensing agreements for their music to train GAI models with AI companies like Suno and Udio.
Examples of other non-textual work AI licensing agreements include:
Collective Licensing Provides a Scalable Solution
While some copyright owners are able to directly negotiate agreements for the use of their works with AI companies, other independent creators and individual copyright owners, who may find direct licensing to be less feasible, often engage in collective licensing. Historically, collective licensing has provided the solution to issues of scalability, and provides the same solution in the GAI context. The Copyright Clearance Center (CCC) was founded in 1978 with the aim of facilitating photocopying permissions in academic settings, and it’s been undeniably successful at distributing royalties at scale. Similarly, Performing Rights Organizations (PROs) collect and distribute monies for musicians where it would otherwise be difficult or inefficient to directly license public performance permissions.
In the GAI context, there are a growing number of collective licensing solutions to address the tremendous scale of GAI training. CCC is an example of a company that evolved to create innovative collective licensing solutions for various types of organizations, including businesses and academic institutions, and for a diverse range of applications, including the use of copyrighted materials with AI systems. CCC’s Annual Copyright License offers reuse rights licensing for copyright owners to easily license their works in a streamlined approach. Created by Humans is a collective licensing platform dedicated to authors of books and other literary works, and Pro Sound Effects developed as a private dataset of audio for flexible licensing. ProRata.ai prides itself as being the first ethical AI search engine and has partnered with more than 500 leading publications providing attribution and revenue shares with all licensors. Protégé acts as an intermediary between AI companies and rights holders that seek to license their audiovisual works for AI training. Getty Images permits legally protected AI image generation by compensating creators for the use of their visual works in AI models. These are just a few of the many options in the GAI licensing market for copyrighted works, and more are being launched to satisfy the needs of users without compromising copyright protections.
Licensing Market Harm Cannot Be Ignored
Copyright law is the framework within which rightsholders have the choice to license their works. Recent court cases Bartz v. Anthropic and Kadrey v. Meta make fatal mistakes in their fair use analyses by ignoring mounting evidence of licensing markets to incorrectly decide that there is no harm to actual or potential licensing markets for AI training, or that copyright owners are not “legally entitled” to these markets. Those determinations misapply the fourth fair use factor’s requirement that the effect of a use on existing markets must be considered. There is no denying that licensing markets for AI training already exist and continue to grow, and to dismiss them as something a copyright owner is not “entitled” to is a misinterpretation of the law that strips copyright owners of their rights.
Ignoring licensing market harm results in damaging economic and legal consequences. Finding that AI licensing is not a market copyright owners are entitled to, courts in the Northern District of California threaten to destabilize the AI licensing market and undermine the entire copyright licensing structure that incentivizes creators. It is inconsistent with established fourth factor precedent to simply conclude that lost licensing fees don’t constitute market harm because such a market may not have been cognizable when a work was created.
The benefits of copyright and GAI can, and should, grow together. Strong IP law results in strong AI; to innovate, prior innovation must be protected and the incentive structure of copyright law must be upheld. For a use to be considered “fair use” when it erases a $2.5 billion market that has even further potential to grow is short-sighted and will ultimately harm both copyright owners and those who wish to use their works for greater technological development.
Conclusion
The AI licensing market for copyrighted works exists, is growing, and is rapidly diversifying to account for the needs of both rightsholders and AI companies. AI companies claiming they do not need a license, and any judges who agree with them, completely ignore, and thus prematurely undermine a nascent, but burgeoning market. Other more responsible AI companies like Bria.ai understand that GAI development and robust licensing for creative works go hand-in-hand with a future where GAI technologies are sustainable. The goal of licensing is to protect rightsholders and create economic prosperity through legal certainty. Negating an active market risks undermining the objective of a healthy and secure copyright system and the incentives afforded by copyright law for the creation and distribution of new works for the public to enjoy.
If you aren’t already a member of the Copyright Alliance, you can join today by completing our Individual Creator Members membership form! Members gain access to monthly newsletters, educational webinars, and so much more — all for free!