Court filings show Meta paused efforts to license books for AI training

Recent court documents in a legal battle over AI copyright involving Meta reveal that the company halted negotiations with book publishers regarding licensing deals for training data for its generative AI models. This pause in discussions aligns with previous reports of the company’s decision to temporarily suspend such efforts.

The case in question, Kadrey v. Meta Platforms, is just one of many similar cases in the U.S. legal system where AI companies clash with authors and other intellectual property owners. The crux of the conflict lies in whether training on copyrighted content constitutes “fair use,” with AI companies typically arguing for this defense while copyright holders vehemently oppose it.

According to the new court filings submitted on Friday, excerpts from depositions of Meta employees indicate that some staff members doubted the scalability of negotiating AI training data licenses with book publishers. Sy Choudhury, head of Meta’s AI partnership initiatives, mentioned encountering limited engagement and interest from publishers during their outreach efforts.

Choudhury explained that despite reaching out to numerous publishers, they faced challenges in establishing contact and receiving feedback, with only a few showing interest. As a result, Meta decided to pause certain AI-related book licensing endeavors due to timing and logistical obstacles, particularly discovering that some fiction book publishers did not actually hold the rights to the content Meta sought to license.

The plaintiffs, represented by notable authors Sarah Silverman and Ta-Nehisi Coates, have continuously revised their complaint since initiating the case in 2023. Their latest complaint accuses Meta of referencing pirated books alongside copyrighted ones to evaluate potential licensing agreements and of utilizing “shadow libraries” containing pirated e-books to train their AI models, including the Llama series.

Allegations suggest that Meta may have accessed these libraries through torrenting, a method that involves uploading files while downloading them, which the plaintiffs argue constitutes copyright infringement.