On March 30, 2026, the Authors Guild (AG), the Association of American Publishers (AAP), News/Media Alliance (N/MA), and the International Association of Scientific, Technical & Medical Publishers (STM) filed an amicus brief supporting the plaintiffs in Concord Music Group v. Anthropic. This case was brought in October 2023 by several music publishers alleging that Anthropic unlawfully used copyrighted musical works, particularly a large corpus of song lyrics, for training the AI product Claude. The case is before the United States District Court for the Northern District of California. 

The joint amicus brief explains that copyright law does not permit Anthropic, a multibillion-dollar company, to systematically copy human-authored works without permission, let alone to enrich itself by generating content that displaces the works it has taken. On fair use factor one, the amici highlight the latest academic research showing that large language models like Claude memorize works used in training in a manner that cannot be “transformative.” 

Even more importantly, on fair use factor four, the brief explains how unauthorized training hurts multiple markets for publishers’ valuable content—including providing direct substitutes to readers and undercutting the now-established markets for licensing AI data. As the brief explains, transformative use is a misplaced argument under the facts of this case and would not, in any event, overcome the “copyright owners’ right to exploit established, valuable markets for their works, including the AI training and [retrieval-augmented generation] RAG markets.” 

The following is a joint statement from Maria A. Pallante, AAP President and CEO; Danielle Coffey, N/MA President and CEO; Caroline Sutton, STM CEO; and Mary Rasenberger, AG CEO:

“This case illuminates the critical, collaborative licensing markets that are developing among copyright owners and technology companies for consumer-facing AI products, driving better, safer, and fairer outcomes for all involved. These partnerships are clearly in the public’s interest, but they will not be fully realized if categorical fair use arguments are permitted to overtake the equities and promise of the Copyright Act.”  

Excerpts from the brief: 

• Given the widespread adoption of AI-related products throughout virtually all aspects of society, it is clear that the training market for AI systems that depend upon copyrighted works for their very existence is one in which copyright owners would naturally participate—and are participating.
• Anthropic, a company currently valued at 380 billion dollars, asks this Court to excuse as a fair use massive, commercially motivated copying of song lyrics to enable its large language model (“LLM”) Claude to generate substitutional works—including nearly identical lyrics—a ruling that would have disastrous consequences for the music plaintiffs in this action and the songwriters whose works they invest in and publish.
• Even beyond the confines of this case, such a ruling could also eviscerate a vital market for authors and publishers of books, news and magazines, scholarly articles, and other textual works who license those works to AI companies to train and operate their LLMs.
• Anthropic’s peers have obtained licenses to use textual works to develop and operate AI systems. Anthropic, however, seems committed to being a holdout by refusing to share with human authors any of the enormous value it reaps from the exploitation of their works.
• A healthy and incentivized licensing market advances the goals of copyright by permitting AI companies like Anthropic to access and use high-quality materials to train better, more sophisticated AI models while protecting rightsholders from uncompensated exploitation.
• Amici’s members’ production of reliable, original news content, literature, scholarship, and other textual works is dependent upon their ability to recoup their investment in quality journalism and authorship. Without copyright protection—and the ability to earn revenue—there would be no sustainable business model for the creation and distribution of creative works.
• Contrary to AI companies’ claims that their copying is transformative because copied works are not retained in their LLMs, a recent study by Stanford and Yale researchers confirms that textual works do not disappear once ingested into the models, but instead are memorized by the system and can be reproduced as output.
• The findings of this study mirror those of others concluding that AI systems are not simply “learning” statistical information about works, but storing the works themselves. In short, memorization of copyrighted texts by LLMs like Anthropic’s Claude is far more significant “than previously understood.”
• Healthy licensing markets advance the core objectives of copyright law by encouraging both the creation and dissemination of expressive works while facilitating efficient and sustainable technological innovation. 

The full amicus brief can be found here.