Authors accuse Meta of using pirated books for AI training

Meta Platforms META.O used pirated books to train its artificial intelligence systems with CEO Mark Zuckerberg’s approval, a group of authors alleged in newly disclosed court documents.

Authors, including Ta-Nehisi Coates and comedian Sarah Silverman, made the accusations on Wednesday in a federal California court. They said Meta’s internal documents showed the company knowingly used pirated works during AI training.

Meta spokespersons did not immediately respond to requests for comment.

The authors first sued Meta in 2023, alleging their books were misused to train Meta’s language model, Llama. The case is one of several alleging AI companies used copyrighted works without permission, with defendants claiming fair use.

The writers asked the court Wednesday for permission to file an updated complaint. They said new evidence revealed Meta trained its AI using the LibGen dataset, a repository allegedly containing millions of pirated works distributed via torrents.

Internal Meta communications showed Zuckerberg approved using LibGen despite concerns from Meta’s AI executives that it contained pirated material, according to the authors.

Last year, U.S. District Judge Vince Chhabria dismissed claims that Meta’s chatbot outputs infringed copyrights and that Meta removed copyright management information (CMI) from the authors’ books.

The writers argued Wednesday that the evidence bolsters their claims, warranting the revival of their CMI claim and adding a new fraud allegation.

During a hearing Thursday, Chhabria allowed the authors to amend their complaint but expressed doubts about the fraud and CMI claims’ viability.