US judge approves $1.5bn Anthropic settlement over pirated book training

A federal judge in San Francisco has granted final approval to a $1.5 billion settlement between Anthropic and a class of authors. The agreement resolves accusations that the artificial intelligence company trained its Claude models on millions of titles downloaded from shadow libraries.

Judge Araceli Martínez-Olguín signed the order on 20 July, closing the largest copyright recovery in United States history. The payout covers roughly half a million works and averages about $3,000 per book.

For European publishers and technology investors, the sheer scale of the settlement serves as a stark financial warning. While the case is bound by US law, the precedent of massive liability for unlawful data scraping is already reshaping risk assessments for artificial intelligence firms operating globally.

The resolution follows a pivotal June 2025 ruling by Judge William Alsup, who determined that training models on lawfully purchased books constitutes fair use. However, he found that acquiring data from sites like LibGen and the Pirate Library Mirror crossed the line into infringement.

Facing potential statutory damages that could have reached hundreds of billions of dollars, Anthropic opted to settle in September 2025 rather than face a jury. The company has been keen to emphasise that the agreement establishes no binding legal precedent.

“We reached this settlement in 2025, after the court’s landmark ruling that training AI on books is fair use under copyright law, which remains the law today,” said Aparna Sridhar, Anthropic’s deputy general counsel. This outcome allows the firm to retain its fair-use victory while paying for the illicit library it downloaded.

Judge Martínez-Olguín initially hesitated, demanding greater clarity on legal fees and compensation structures before granting approval. She ultimately reduced the attorneys’ fee award by approximately $86 million from the original request to ensure the deal fairly compensated the authors.

The original 2024 class action was led by novelists Andrea Bartz and Charles Graeber alongside non-fiction writer Kirk Wallace Johnson. Their complaint accused the company of having “attempted to steal the fire of Prometheus” by “strip-mining the human expression and ingenuity behind each one of those works.”

With final approval secured, disbursement to eligible rightsholders can now begin. According to the Authors Guild, the standard trade split will direct half of each book’s payment to the publisher and half to the author, barring alternative contractual terms.

The wider technology sector will note the $1.5 billion figure closely as comparable disputes involving Meta, Google, and OpenAI continue to grind through the courts. None of those ongoing cases have yet produced the tidy legal distinction between lawful training and unlawful acquisition seen in this ruling.

Consequently, the next defendant to face a jury may find it much harder to negotiate a contained settlement. Anthropic ultimately secures the fair-use framework it desired, but at a steep price for the shadow library it chose to download.