A collected series of cases captioned as In re OpenAI Copyright Infringement Litigation presents the latest and most important opportunity to resolve an emerging and critical question in copyright law: does using copyrighted materials to train AI infringe those copyrights?
The cases involve OpenAI’s use of copyrighted material to train its large language models (LLMs), including material from news sources and books. Sources include the New York Times, multiple Daily News newspapers, Ziff Davis, the Center for Investigative Reporting, and The Intercept Media , and from book authors along with the Author’s Guild, collectively known here as “Plaintiffs.” The facts are clear that OpenAI used these rightholders’ copyrighted materials to train its LLMs without permission. However, in certain circumstances, the use of copyrighted material is not infringing as it rests under the legal doctrine of “fair use.” Thus, the key issue for the court is whether OpenAI’s use qualifies as fair use. If so, then OpenAI’s use was permissible and not infringing at all. If not, then OpenAI could be liable to the rightsholders and may be required to pay licensing fees for future uses of the materials, an outcome that could affect, delay, or restrict the development of AI tools and models.
CCIA, along with our fellow amici curiae, submitted a brief to urge the Court to find that OpenAI’s use qualifies as fair use. Our brief makes three key points: (1) The purpose of the Copyright Act, and of fair use in particular, is to promote progress—not protectionism; (2) training generative AI models is a transformative use; (3) Plaintiffs’ market dilution and licensing theories misconstrue the fair use inquiry and are antithetical to the purposes of copyright.
The purpose of the Copyright Act, and of fair use in particular, is to promote progress—not protectionism.
As we assert, the claims against OpenAI must be viewed according to the purpose of copyright law, namely to promote technological progress and public benefit, and not to shield incumbent rightsholders from competition. This purpose arises from the Constitution, which enables Congress to “[p]romote the Progress of Science and useful Arts, a power Congress exercised when it passed the Copyright Act. Indeed, as the Supreme Court clarified in Twentieth Century Music Corp. v. Aiken, “[c]reative work is to be encouraged and rewarded, but private motivation must ultimately serve the cause of promoting broad public availability of literature, music, and the other arts.” This cultural progress occurs through an ever-more-sophisticated series of tools. Just as cameras, word processors, and digital editing software did not replace human creators, generative AI models serve as sophisticated instruments that assist human users—whether researchers synthesizing scientific literature, clinicians reviewing medical records, or authors drafting new works—who remain the driving force behind the creative process, the results of which obtain limited protections sufficient to enable even greater progress. Copyright should serve this progress and not protect rightsholders at the expense of potential greater advancements.
Training generative AI models is a transformative use.
In defining “fair use,” the Copyright Act specifies 4 factors to consider:
- the purpose and character of the use, including whether such use is of a commercial nature or is for nonprofit educational purposes;
- the nature of the copyrighted work;
- the amount and substantiality of the portion used in relation to the copyrighted work as a whole; and
- the effect of the use upon the potential market for or value of the copyrighted work.
In evaluating the first factor – the purpose and character of the use – courts often ask whether the use of the copyrighted material is “transformative,” meaning whether the use “adds something new, with a further purpose or different character, altering the first with new expression, meaning, or message.” As noted in Campbell v. Acuff-Rose Music, Inc., the “more transformative the new work, the less will be the significance of other factors, like commercialism, that may weigh against a finding of fair use.”
Applying the first fair use factor, it’s plain that generative AI training is among the most clearly transformative uses a court has ever evaluated. Developers do not copy books and articles to read, display, or distribute their expressive content to the public. Instead, they engage in intermediate, non-expressive copying for a purely computational purpose: extracting statistical patterns and word relationships across billions of text fragments to build fundamentally new tools. Consistent with Supreme Court precedent in Google v. Oracle and Second Circuit rulings in Google Books and HathiTrust, copying entire works to build innovative and transformative information technologies is squarely protected by fair use.
The Plaintiffs’ market dilution and licensing theories misconstrue the fair use inquiry and are antithetical to the purposes of copyright.
Our brief also responds to Plaintiffs’ flawed theories of market harm under the fourth fair use factor: “market dilution” and lost licensing revenue. “Market dilution” refers to the argument that AI might create competing works that flood a market and indirectly substitute for copyrighted works. But copyright law protects only against the usurpation of an author’s specific original expression through substantially similar substitutes; it has never granted a monopoly against general competition from new, non-infringing works, shared genres, or the reporting of uncopyrightable facts. Recognizing a “market dilution” theory would stretch copyright beyond its constitutional limits and raise serious First Amendment concerns by penalizing tools that expand the creation of new speech.
Similarly, rightsholders cannot manufacture cognizable market harm through circular reasoning merely by demanding licensing fees for a transformative use. While some AI developers enter into voluntary agreements for convenience, data formatting, access to paywalled archives, or litigation avoidance, those business arrangements do not strip the broader technology sector of its statutory right to make fair use of information.
Ultimately, the rule the Court adopts here will reverberate far beyond the parties in this case, directly impacting startups, university labs, non-profit researchers, and open-weight model developers who rely on fair use to innovate. Models trained on broad collections of works are being adapted to analyze genomic data, translate low-resource languages, detect fraud, and power assistive technologies for people with disabilities. Allowing incumbent rightsholders to lock up training data would stifle this vast downstream innovation and creativity. CCIA and its fellow amici have therefore asked the Court to join the recent decisions in Bartz v. Anthropic and Kadrey v. Meta and hold that training generative AI models on copyrighted works is fair use.
