What do authors John Grisham and Ta-Nehisi Coates, performers Billie Eilish and Stevie Wonder, visual artist Sarah Andersen, and corporations like Getty Images, the New York Times, and Sony Music, all have in common? (Along with a long tail of others you probably have never heard of, such as writers Andrea Bartz, Charles Graeber and Kirk Wallace Johnson.) They belong to an army of creators and publishers who have recently claimed catastrophic harm from generative-AI models like Anthropic’s Claude and OpenAI’s ChatGPT.
Today, controversies surrounding AI are increasing on a daily basis, and at the center of the storm are questions about intellectual property rights (IPRs). Ever since the first printing press, the same confrontation has continually occurred between disruptive new technologies and existing copyright holders. The publishing industry sanctimoniously wails that the new innovation will assuredly mean the death of creativity and authorship. And their lobbying for legislative barriers invariably imposes constraints on consumer welfare and derivative rights. Meanwhile, publishers successfully alter their existing business models (to the extent that revenues in the music industry have actually increased over time.)
ChatGPT, Suno, and Claude belong to a class of rapidly-evolving prediction machines. Claude’s large language model (LLM) predicts text, while Suno’s neural networks predict music sequences. Clearview AI uses billions of photographs scraped from social media platforms, which generates facial recognition even from partial features. Such AI models are trained on vast amounts of material from the internet without attention to their copyright status, allegedly perpetrating illegal infringement on a massive scale.
AI models are further capable of producing creative and original works in art and technology, and the U.S. Patent Office and Copyright Office must consider whether AI-generated ideas and expression should be granted patents and copyrights on their own account. At present, in the legal world, AI entities are analogous to monkeys like Naruto (selfie pictured here), whose works are relegated to the public domain as they are not deemed human enough to be granted IPRs.
Other countries are grappling with the same issues. For instance, Getty Images is arguing before the UK High Court that Stability AI illegally used millions of its copyrighted images to train its AI model. The House of Lords is prioritizing the rights of “creatives” and has rejected legislative attempts to reach a compromise solution that mediates the claims of all parties. However, international policies towards copyright hold limited lessons for the U.S.
Fair Use or AI Piracy?
My books and research in law and economics empirically assess two centuries of U.S. intellectual property policies, and the results support the argument that AI should not be treated as rampant pirates.
The U.S. approach to copyrights has always been uniquely utilitarian. According to the U.S. Constitution, the objective of copyright is to “promote the progress of science and useful arts.” As Congress pointed out in 1909, this means U.S. copyright is granted “not primarily for the benefit of the author, but primarily for the benefit of the public…” Courts have the primary objective of advancing social welfare, while recognizing the rights of copyright holders. This “balance on the ledger” came down heavily on the side of international copyright piracy for the first century of American history, and my research shows that weak copyrights in this era increased economic progress and social welfare.
The legal doctrine of “fair use” allows free unauthorized access to copyrighted material, and considers “four factors.” A primary consideration is whether the use was “transformative” or sufficiently different from the original works. Noncommercial and educational uses are favoured by the courts over for-profit endeavours. It is likely not coincidental that OpenAI has opted to remain non-profit and is pushing for widespread educational adoption of ChatGPT, since both tilt the balance towards permissible fair use.
The most important factor comprises the effect of unauthorized use on the market for the copyrighted work. The recent landmark decision in Thomson Reuters v. Ross Intelligence (2025) found that the defendant violated the law because its AI usage was commercial and untransformative, and further cannibalized the market for Westlaw’s services.
Send in the Economists
The copyright question of market impact essentially involves economic thinking.
A key question that has not been adequately considered is, what is the relevant market? Do prompts on ChatGPT provide a substitute for readership of daily news which expropriates revenues that would otherwise go the New York Times? Here, it is important to distinguish between input markets and output markets. The central complaint in these lawsuits is that LLMs are being trained on big data comprising copyrighted material, and that is certainly correct. However, LLMs use copyrighted materials as an input to generate predictions and, from an economic perspective, this input market is distinct and separate from the market for copyrighted output. Any profits AI firms earn derive from their unique market for prediction, rather than from lost revenues to copyright owners in the market for content.
In some instances, “piracy” can actually benefit the market for the original work, such as when a ChatGPT response leads the prompter to purchase a copy of a book mentioned. Relatively unknown authors (or academics like myself) might welcome free use of their works that potentially adds to their reputation. Piracy can also promote network expansion, scalability and superstar effects. Taylor Swift and Beyonce give away their music on social media like YouTube, to build up a network of supporters which they subsequently monetize through performances, merchandise, and complementary forms of commercialization.
When is Copyright an Unjust “Tax on the Public”?
At the start of the digital music revolution, the Recording Industry Association of America and music publishers sued Napster, Grokster and other file-sharing platforms. Appellate courts rejected the fair use defense, which ultimately led to the demise of Napster and contemporary peer-to-peer music providers. However, when Sony came out with the Betamax VCR recorder, they claimed fair use of Universal’s copyrighted works, and the Supreme Court agreed. Ironically, today Sony is on the other side of the battle, accusing AI technologies of illicit piracy.
Digital technology makes it possible for many users to simultaneously gain access to the same work without depriving the original owner. Twenty years ago, Google boldly launched its initiative to create digital versions of millions of books, without the permission of copyright holders. The Authors Guild disputed this unauthorized use but, despite the scale and amount of copying involved, the federal courts rejected their arguments in 2015.
Current AI copyright litigation is based on a zero-sum game assumption that gains to one group necessarily imply losses to another. The legions of plaintiffs today are claiming statutory damages from AI firms for unauthorized use of billions of copyrighted materials, which could lead to penalties of $150,000 per willful violation. Will the Google or Napster approach prevail? Just as in the case of Napster, a landmark appellate decision against fair use could herald the start of another “AI winter” — and that would hardly serve to promote the progress of science and the useful arts.
