A federal judge has approved a $1.5 billion settlement between artificial intelligence firm Anthropic and a class of authors and publishers, the largest copyright class action settlement in United States history. Judge Araceli Martinez-Olguin signed the approval on July 20, 2026, in the case known as Bartz v. Anthropic, resolving claims that the company built its Claude models on books it took without permission.
The writers who brought the case sued in 2024. They alleged that Anthropic copied their books from pirate libraries and fed them into the training data behind Claude, its family of AI models. As TechCrunch reported, the approval turns a long running fight over how AI systems learn into a concrete bill, and it hands writers a rare win in a wave of litigation that has mostly moved slowly.
Under the terms, authors and publishers will receive about $3,000 for each of roughly 500,000 works the company copied. Courthouse News reported that the opt in rate reached about 92.77 percent, with 447,576 works claimed by the time the deal neared final approval. That level of participation runs high for a class action of this size, and it signals that most eligible writers preferred a guaranteed payout to the uncertainty of pressing their own suits.
Anthropic will pay the money in four installments. The first $300 million came due by October 2, 2025. Another $300 million lands within a week of final approval. A further $450 million is set for September 25, 2026, and the final $450 million is due by September 25, 2027. Spreading the payments across roughly two years softens the immediate hit for a company that has raised heavily from investors but still spends aggressively on model training and compute.
The distinction that matters most sits in what the settlement does not do. A settlement is not a court ruling that training AI on copyrighted work is illegal. Earlier in the case, the court had weighed whether using books to train a model could count as fair use, a question that remains unsettled across the wider legal landscape. What created Anthropic’s exposure was not the act of training itself but the way it obtained the books. Copying works from pirate libraries left the company open to piracy claims, and that piracy, rather than the training, is what drove it to the table.
For the AI industry, the number carries the message. Companies including OpenAI, Google, Meta, and Anthropic have trained large models on enormous piles of text, and the provenance of that text has become a legal soft spot. The Anthropic deal puts a price on one version of the problem. If a firm sources copyrighted material from pirated collections, it now has a rough sense of what a class of authors might extract in return. That figure will shape how legal teams weigh the cost of scraping against the cost of licensing.
For writers, the outcome reads as validation, though a narrow one. The Authors Guild, which has tracked the case closely, framed the settlement as a meaningful recovery for members whose books appeared in the training sets. Yet the money flows only to those who filed claims for identified works, and the deal leaves the broader question of lawful training for another court on another day. Writers who hoped for a definitive judgment that AI training requires consent did not get one here.
The case also lands at a moment when licensing deals between AI companies and publishers have started to multiply. Several model builders have signed content agreements with news organizations and book publishers, moving toward paid access rather than open scraping. The Anthropic settlement gives those negotiations a sharper backdrop. A publisher weighing a licensing offer can now point to the cost of getting it wrong.
Anthropic has not framed the payment as an admission that its training methods break the law, and the settlement avoids any such finding. The company gets to close a costly and distracting piece of litigation, keep its models running, and move on with a clearer view of its liabilities. The authors get paid. The larger fight over what AI systems may read, and under what terms, continues in other courtrooms.
*Disclosure: Entrelligence uses Anthropic’s Claude tooling in its editorial workflow.*
That leaves a financial precedent more than a legal one. The price of training on pirated books now carries a public number, and the AI labs, authors, and publishers still fighting similar cases will negotiate with that figure in front of them.
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →