Anthropic paga $1.500M por piratería de libros pero evita jurisprudencia decisiva

A federal judge in California signed off on a settlement this week that will cost Anthropic $1.5 billion—the largest copyright payout in American history. The company admitted it had downloaded roughly half a million books without permission from piracy platforms like Library Genesis to feed its Claude language model. Yet for all the money changing hands, the settlement sidesteps the fundamental legal question that will shape how artificial intelligence companies can use copyrighted material.

The case began in 2024 when writers and publishers sued Anthropic for copyright infringement. Judge Araceli Martínez-Olguín approved the settlement on behalf of the plaintiffs, calculating damages at $3,000 per work—four times the minimum statutory penalty under U.S. intellectual property law. The judge noted that continuing the trial would have been protracted, expensive, and risky for the authors and editors. An unfavorable verdict at trial could have left them with no recourse.

The settlement's approval marks a turning point in how the tech industry faces accountability for its data practices, but it also represents a strategic retreat from the courtroom. The original judge overseeing the case, William Alsup, had signaled that while training AI on legally obtained books might qualify as fair use, downloading and permanently storing files from pirate sites constituted clear copyright violation. Alsup retired at the end of 2025, but his reasoning proved decisive in pushing both sides toward a deal. Anthropic chose to pay rather than risk a precedent-setting ruling that could have constrained how it and other companies train their models.

The company's legal team framed the settlement as validation. Aparna Sridhar, Anthropic's deputy general counsel, called it confirmation that training AI on books constitutes legitimate use under copyright law. She noted that over 91 percent of covered authors and publishers claimed their share of the payout. The Authors Guild and major publishers hailed the outcome as a crucial defense of intellectual property rights against tech corporations. Justin Nelson, the plaintiffs' attorney, called it the largest copyright settlement ever achieved.

But experts see the real significance elsewhere. Beatriz Busaniche, director of the Vía Libre Foundation, points out that dozens of similar lawsuits are pending across the United States and Europe. The Anthropic settlement, she argues, functions like a private mediation—the parties reached an agreement and moved on, leaving the central legal question unresolved. Other cases will continue until a court ruling establishes binding precedent. U.S. copyright specialists predict those cases could take five years or more to reach the Supreme Court, an eternity in technology. Anthropic essentially bought its way out of a decision that might have gone against the entire industry.

Meta and OpenAI face comparable lawsuits. The stakes are whether training AI systems on copyrighted material that remains under active copyright protection violates authors' rights. Preliminary rulings so far have suggested that training on legally obtained books does not infringe copyright. For authors, allowing that logic to solidify into precedent would be catastrophic. The settlement with Anthropic, from their perspective, is pure gain—compensation without the risk of a ruling that could legitimize what the company did.

Magdalena Iraizoz, director of Cadra, a collective management organization, sees the agreement as a necessary step toward a broader solution. She draws a parallel to the photocopier era, when companies eventually agreed to compensate authors for reproduction. The same principle should apply to AI training, she argues, though it remains unregulated. Collective management entities worldwide will eventually become essential to handling licensing at scale, ensuring that creators receive payment when their work trains artificial intelligence systems.

The settlement validates that Anthropic wronged authors and must pay. It does not, however, answer whether the company's core practice—using copyrighted material to build AI—is legal or illegal. That question remains open, waiting for a court willing to answer it.