OpenAI has shared excerpts from court briefs submitted by a broad group of authors, journalists, creators, academics, librarians, accessibility advocates, legal experts, economists and business leaders. The submissions support the application of fair use to AI training in litigation brought by The New York Times.
The central argument across the briefs is that AI systems analyze copyrighted works to learn statistical patterns rather than reproduce or distribute those works. Supporters say this makes training a transformative use that serves a different purpose from the original material and does not provide the public with substitute access to it.
Several submissions argue that copyright should protect authors against unauthorized substitution for their protected expression, but not against competition from new works that share a genre, subject matter or audience. Authors Alliance described claims of broad “market dilution” as an attempt to make copyright cover competitive pressure that the law has historically left outside rightsholders’ control.
Organizations representing libraries, accessibility advocates and technology companies warned that restrictions on AI training could affect more than the companies developing large language models. They said such limits could also influence medical research, education, accessibility tools, agriculture, independent creators and industries that rely on open-source or third-party AI systems.
Other briefs focused on the practical uses of generative AI. Musicians, writers and other independent creators said AI can reduce administrative costs, help develop ideas, expand creative output and allow smaller businesses to compete with better-resourced rivals. One musician and one comedian-writer described using AI to manage business operations and devote more time to creative work.
Legal scholars and technology groups also rejected the idea that trained models function as searchable repositories of their source material. Computer scientists cited in the filings distinguished generalization from memorization, arguing that models typically generate responses by predicting tokens from learned patterns rather than retrieving stored documents through an address or file system.
The excerpts further point to earlier applications of fair use involving search engines, full-text book search, accessible digital books and the publication of technical standards incorporated into law. Public Knowledge argued that the doctrine can similarly help ensure that the benefits of AI reach the broader public rather than being controlled by a limited group.
Former members of Congress Rick Santorum and Rick Boucher described fair use as a principle intended to operate alongside technological change. They characterized AI training as a significant test of that principle and said the ability of scientists, companies and entrepreneurs to develop AI has economic and strategic importance.
The United States government, according to the published excerpt, also opposed narrowing fair use to exclude the training of OpenAI’s large language models. The full briefs are available through the public court docket, while OpenAI’s page provides selected passages supporting its position that fair use promotes innovation and public benefits.
Sources: OpenAI News.