Federal Judge Allows Authors’ AI Copyright Case Against Meta

A federal judge allows an AI copyright lawsuit against Meta to move forward, though he dismissed a portion of the original suit.

A group of authors, including Richard Kadrey, Ta-Nehisi Coates, and comedian Sarah Silverman, filed a lawsuit against Meta with allegations that the company violated copyright by training its AI models using their books. Meta has argued that training its LLMs qualifies as fair use, asking the judge to dismiss the lawsuit outright.

In a ruling on Friday, US District Judge Vince Chhabria determined that the authors’ allegation of copyright infringement is “obviously a concrete injury sufficient for standing.” Perhaps even more damning, the judge also ruled that the authors “adequately alleged that Meta intentionally removed CMI [copyright management information] to conceal copyright infringement.”

“These allegations raise a reasonable, if not particularly strong, inference, that Meta removed CMI to try to prevent Llama [Meta’s AI model] from outputting CMI and thus revealing it was trained on copyrighted material,” wrote Judge Chhabria.

Notably, the judge dismissed the authors’ claims that Meta violated the California Comprehensive Computer Data Access and Fraud Act (CDAFA). He explains this is due to the fact they did not “allege that Meta accessed their computers or servers — only their data (in the form of their books).”

Even with part of the lawsuit dismissed, things don’t look good for Meta. Discovery in the case has shown that Meta employees discussed internally the use of copyrighted works to train the company’s LLMs. Documents submitted to the court allege that Meta CEO Mark Zuckerberg gave his AI team the go-ahead to train on copyrighted content acquired through “legally questionable” means, while halting AI training data licensing talks with book publishers.

One Meta employee, a research engineer named Xavier Martinet, suggested in internal chats that the company buy ebooks at retail prices to build a training set rather than securing licensing deals with individual book publishers. When another employee pointed out that this could put them in a position where they could be legally challenged, Martinet doubled down on asking for forgiveness rather than permission.

“I mean, worst case: we found out it is finally ok, while a gazillion start up [sic] just pirated tons of books on bittorrent,” wrote Martinet. “My 2 cents again: trying to have deals with publishers directly takes a long time…”

Even Melanie Kambadur, a senior manager for Meta’s Llama model research team, suggested using “links aggregator” LibGen to provide access to copyrighted works from publishers, rather than licensing data sources individually. LibGen has been sued several times and fined millions for copyright infringement.