Recently, Pulitzer Prize-winning author John Carreyrou and other writers filed a copyright lawsuit accusing six AI giants of using pirated copies of their works to train large language models, constituting “deliberate theft.”

The complaint was filed in the U.S. District Court for the Northern District of California, naming Anthropic, Google, OpenAI, Meta, xAI, and Perplexity AI as defendants. This marks the first copyright lawsuit against xAI over its training practices and the first time Perplexity has been sued by authors.

xAI responded via email: “Traditional media is lying.” Jesse Dwyer, Perplexity's communications director, stated the company does not index books. The other defendants have not yet responded.

Plaintiffs allege these AI companies downloaded pirated copies of their works from “illegal shadow libraries” like LibGen, Z-Library, and OceanofPDF. The defendants' illegal downloading of books constitutes primary infringement, while creating additional copies during model training or “optimizing” products constitutes secondary infringement.

The authors assert their works “now underpin a multibillion-dollar product ecosystem” without compensation.

Plaintiff Carrelo previously opted out of a $1.5 billion class action settlement with Anthropic. One of the law firms representing the authors, Freedman Normand Friedland LLP, appeared in court last month to defend another firm accused of attempting to “trick” authors into opting out of the Anthropic settlement. The judge overseeing the settlement warned that firm against promising authors they could sue Anthropic individually only to renege later, stating that “would be highly unethical conduct, sufficient to disbar you.”

All plaintiffs attended the November hearing to assure the court they had not been deceived into opting out of the settlement.

The authors are not pursuing this case as a class action. The complaint states this is because they do not wish to “settle claims for a pittance.” They cite the example of the aforementioned Anthropic settlement agreement, under which each work would be entitled to only about $3,000. If a jury finds the infringement to be willful, authors could receive up to $150,000 in damages per infringed work.

The complaint asserts: “Large language model companies should not be able to easily offset thousands of high-value claims with such a low floor, thereby evading the true costs they should bear for their massive, willful infringement.”

Scan the QR code to access the complaint.