Pulitzer Winner Leads Copyright Revolt Against AI Giants, Targets Musk's xAI

Six prominent authors including John Carreyrou are abandoning class-action settlements to pursue individual lawsuits worth up to $150,000 per book against major AI companies.

John Carreyrou spent years exposing Silicon Valley fraud at Theranos. Now the Pulitzer Prize winner is targeting what he calls a different kind of theft: AI companies training their models on pirated copies of his books. On December 22, Carreyrou and five other authors filed suit against OpenAI, Google, Meta, Anthropic, Perplexity, and—for the first time in any major copyright case—Elon Musk's xAI.

The lawsuit represents a strategic gamble that could reshape how authors fight back against AI training. Rather than accept the roughly $3,000 per book offered through class-action settlements, these authors are pursuing individual suits seeking statutory damages of up to $150,000 per work. The timing appears deliberate: Judge William Alsup, who had criticized the "pennies on the dollar" settlement in the landmark Anthropic case, took inactive status on January 12, with the case reassigned to Judge Araceli Martinez-Olguin. Other authors have until January 29 to decide whether to follow Carreyrou's path.

According to the complaint filed in Bloomberg Law, the defendants committed a "deliberate act of theft" by using shadow libraries like LibGen and Z-Library to train their models. Attorney Kyle Roche of Freedman Normand Friedland, representing the authors, argued in court that "stealing books to build its AI was Anthropic's 'original sin' and the settlement did not go far enough."

The inclusion of xAI marks an escalation. When asked for comment about the lawsuit, the company responded with a three-word email: "Legacy Media Lies." No formal legal response has been filed.

Perplexity offered the only substantive denial. "Perplexity doesn't index books," Jesse Dwyer, the company's head of communications, told Reuters. The statement leaves open whether Perplexity's retrieval-augmented generation system might access shadow libraries indirectly through third-party sources.

OpenAI, Google, Meta, and Anthropic all declined to comment.

The revolt against class-action settlements has been building for months. In September, Anthropic announced a $1.5 billion settlement fund for the Bartz class action—one of the largest copyright settlements in history. But when divided among an estimated 500,000 works, individual payouts shrink dramatically. Judge Alsup expressed skepticism during a November hearing, questioning whether "interloper" lawyers were adequately representing authors' interests.

"LLM companies should not be able to so easily extinguish thousands upon thousands of high-value claims at bargain-basement rates," Roche wrote in the complaint.

The technical claims mirror those in previous suits but add more specificity. The plaintiffs allege that companies used the "Books3" dataset, a collection of roughly 196,000 pirated books scraped from shadow libraries. They claim to have evidence that models were trained on verbatim text, not statistical patterns alone.

This creates a different risk calculus for AI companies. A class action caps total exposure—even $1.5 billion split among hundreds of thousands provides certainty. Individual suits seeking maximum statutory damages could theoretically reach tens of billions if enough authors opt out and win.

The judge change adds another variable. Martinez-Olguin lacks Alsup's track record of skepticism toward tech companies in copyright cases. Her approach to these novel AI training questions remains untested.

For video AI creators watching these cases: • Training data provenance becomes even more critical as liability frameworks solidify • "Clean room" datasets may command premium pricing if copyright risk increases • Individual creator suits could follow the author playbook if class actions disappoint • Shadow library connections in any training pipeline create maximum legal exposure • Companies may accelerate synthetic data research to avoid copyright entirely

The January 29 deadline for opting out of the Anthropic settlement will reveal whether Carreyrou's strategy triggers a broader exodus. If hundreds of prominent authors follow suit, the economics of AI training could shift dramatically.

Whether courts will accept the "shadow library" theory of liability remains an open question. Previous fair use defenses have focused on transformation and public benefit. But as Roche argues, there's a difference between analyzing books and "stealing them wholesale to build a commercial product."

The three-word response from xAI may prove telling. Not a fair use defense. Not a denial of using the books. Just "Legacy Media Lies"—as if the medium of the accusation matters more than the accusation itself.

Source notes: • Lawsuit filing and "deliberate act of theft" quote → src_bloomberg_filing • First suit naming xAI → src_reuters_xai • xAI email response → src_bloomberg_filing • Perplexity denial → src_reuters_xai • Judge Alsup inactive status and new judge → src_alsup_update • $1.5B settlement and opt-out deadline → src_alsup_update • Kyle Roche quotes → src_reuters_xai, src_bloomberg_filing • $3,000 vs $150,000 damage figures → src_bloomberg_filing, src_reuters_xai • Company response/no-comments → src_reuters_xai