A Case Study
Generative AI companies train their models on millions of books, articles and news pieces. But can these works be used without the consent of their rights holders? This question now sits at the center of hundreds of lawsuits worldwide. In the United States, one of the most closely watched examples has produced both a precedent-setting court ruling and one of the largest copyright settlements in history.
How Did the Dispute Begin?
On 19 August 2024, authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson filed a class action against Anthropic. The claim: the company had obtained millions of copyright-protected books from pirated digital libraries, without authorization, to train its Claude model. The case quickly grew into a large-scale class action covering approximately 500,000 works.
The Turning Point
In June 2025, Federal Judge William Alsup drew a decisive line.
He ruled that training an AI model on lawfully purchased and scanned books could be considered fair use. Downloading millions of pirated books and retaining them in a permanent digital archive, however, he found to constitute clear copyright infringement.
With this ruling, the dispute shifted from a question of liability to one of the amount of damages.
Why Settle Instead of Going to Trial?
The risk was substantial.
US copyright law allows statutory damages of up to $150,000 per work for willful infringement. Applied to roughly 482,000 works, the theoretical damages exposure reached $72 billion.
Facing uncertainty of this magnitude, the parties chose to negotiate rather than wait for a jury verdict.
The Mediation Process
The parties selected retired federal judge Layn Phillips as mediator.
Following preliminary discussions, intensive negotiations held in August 2025 culminated in a binding settlement. The agreement was reached only a few months before trial.
The court subsequently approved the settlement. That more than 91 percent of rights holders filed claims for payment was a striking indicator of the confidence placed in the settlement. The settlement received final court approval only a few days ago, on 20 July 2026.
The Case Closed, But ADR Continued
One of the most notable aspects of this matter is that the settlement did not merely resolve the existing dispute.
The court also approved a mechanism under which any new disagreements arising during implementation would first be handled by a settlement administrator and, if unresolved, decided by a Special Master operating confidentially.
In this way, dispute resolution continued after the case, not through the courts, but within a structured ADR system.
What Does This Case Show?
The Anthropic case shows that even one of the most complex intellectual property disputes of the AI era could be resolved not by a court judgment but through a structured mediation process.
More importantly, the parties did not merely bring the current dispute to an end; they also built a lasting resolution mechanism for disagreements that may arise in the future.
As AI-driven disputes continue to grow, this approach is expected to gain increasing importance not only in copyright matters but also in other commercial disputes arising from data, technology and the digital economy.
