Judge Approves Anthropic’s $1.5 Billion Copyright Settlement With Authors

A federal judge has approved Anthropic’s $1.5 billion settlement with authors and publishers who accused the artificial intelligence company of obtaining pirated copies of their books.

The agreement resolves a major class action case involving more than 480,000 registered works. It is considered the largest known financial settlement in a United States copyright case. Eligible copyright holders are expected to receive about $3,000 for each covered book, depending on final administrative costs and other adjustments.

United States District Judge Araceli Martínez Olguín granted final approval after reviewing objections from people who believed the payments were too small. Some critics argued that Anthropic could have faced much higher damages if the dispute had continued to trial.

The judge concluded that the agreement still offered meaningful and immediate compensation while avoiding the uncertainty, cost, and delay of further litigation.

The Case Focused on How Anthropic Obtained the Books

The lawsuit began in 2024 when authors accused Anthropic of downloading copyrighted books from unauthorized online collections. The material allegedly became part of a central digital library containing millions of books.

The legal dispute later separated two important questions. The first concerned whether copyrighted books could be used to train artificial intelligence systems. The second concerned whether Anthropic had lawfully acquired and stored those books.

In an earlier ruling, the court found that using books to train artificial intelligence models could qualify as fair use because the training process was considered transformative. However, that finding did not protect Anthropic from claims involving books downloaded from piracy websites.

The court found that building and maintaining a permanent library of pirated books raised a separate copyright issue, even when some of those files were not ultimately used for model training. That distinction exposed the company to a trial where possible statutory damages could have reached a much higher amount.

Anthropic agreed to settle before that trial took place.

Legal Fees Were Reduced by the Court

Lawyers representing the authors originally requested $187.5 million in fees. The judge reduced that amount to approximately $101.6 million after deciding that the larger request would give the legal team an excessive share of the settlement.

Rather than calculating the payment as a simple percentage of the full fund, the court considered the hours worked, reasonable legal rates, the complexity of the case, and the financial risk accepted by the attorneys. The reduction leaves more of the settlement fund available for authors and publishers.

More than 91 percent of the eligible works have already been claimed, showing that most affected rights holders have chosen to participate in the agreement.

Anthropic must also remove and destroy the pirated book collections covered by the settlement. This requirement prevents the company from continuing to store the unauthorized files that formed the central issue in the case.

Some Copyright Claims Will Continue Separately

The settlement does not end every copyright dispute involving Anthropic. Several authors and publishers chose not to participate because they believed they could recover more money through individual cases.

Those parties remain free to continue their own lawsuits. Their claims could test how courts calculate damages when copyrighted material is acquired illegally but later used for artificial intelligence development.

The approved agreement also leaves the earlier fair use decision in place. As a result, the case does not establish that artificial intelligence companies must always obtain licences before training models on copyrighted books. Instead, it reinforces a narrower distinction between using material for training and obtaining that material through piracy.

That distinction is likely to influence other copyright cases involving artificial intelligence companies. Future disputes may focus closely on where training data came from, how it was stored, and whether the developer had permission to possess the original copies.