Meta Pauses Book Licensing for AI Training - Latest News

AI Copyright Case Reveals Meta's Licensing Pause

Recent legal documents filed in an ongoing copyright case against Meta bolster previous reports indicating the company temporarily halted discussions with book publishers. These discussions concerned potential licensing agreements intended to provide training data for Meta’s generative AI models.

The Core of the Legal Dispute

The filings pertain to the case Kadrey v. Meta Platforms, one of numerous legal battles currently progressing through the U.S. court system. These cases involve conflicts between AI companies and authors, alongside other intellectual property rights holders.

Generally, AI companies have maintained that utilizing copyrighted material for training purposes constitutes “fair use.” However, copyright holders strongly disagree with this assertion.

Scalability Concerns with Licensing

Newly submitted court documents, including excerpts from depositions of Meta employees, suggest internal concerns regarding the feasibility of scaling AI training data licenses for books.

Sy Choudhury, who oversees AI partnership initiatives at Meta, stated that initial outreach to publishers yielded limited engagement and interest.

Limited Publisher Response

“We compiled an extensive list of top publishers through initial internet research,” Choudhury explained in his deposition. “However, we received minimal contact or feedback from our initial outreach attempts to establish communication.”

He further noted that only a small number of publishers responded positively to their inquiries.

Pause in Licensing Efforts

According to the transcripts, Meta paused specific book licensing initiatives in early April 2023 due to “timing” issues and other logistical challenges.

It was discovered that some publishers, particularly those specializing in fiction, lacked the necessary rights to license the content Meta was interested in.

“Our business development team quickly determined that many publishers we contacted did not actually possess the rights to license the data to us,” Choudhury stated. “Securing rights from all individual authors would have been a lengthy process.”

Previous Licensing Setbacks

Choudhury also indicated that Meta had previously paused licensing efforts in another context.

“We previously attempted to license 3D worlds from various game engine and game developers for our AI research team,” he said. “Similar to the situation with fiction and textbook data, we encountered very little interest in even initiating a conversation. Consequently, we opted to develop our own solution.”

Allegations in the Amended Complaint

The plaintiffs’ counsel, representing authors like Sarah Silverman and Ta-Nehisi Coates, have repeatedly amended their complaint since the case’s filing in the U.S. District Court for the Northern District of California, San Francisco Division in 2023.

The latest complaint alleges that Meta cross-referenced illegally obtained books with copyrighted books available for licensing to assess the viability of pursuing licensing agreements.

Use of "Shadow Libraries" and Torrenting

The complaint further accuses Meta of utilizing “shadow libraries” – repositories of pirated e-books – to train its AI models, including the Llama series.

Plaintiffs allege that Meta may have obtained access to these libraries through torrenting, a file-sharing method that requires users to simultaneously upload files while downloading – a practice the plaintiffs contend constitutes copyright infringement.