The First Settlement in a GenAI Training Case Has Implications for Content Creators and Business Users

The first-ever monetary settlement in a generative AI (“GenAI”) copyright infringement case was recently announced.

Anthropic, maker of the Claude GenAI, agreed to pay at least $1.5 billion, reportedly the biggest settlement ever for a U.S. copyright infringement case. The settlement concerns the use of pirate databases – places where illegally made copies of books can be downloaded.

The settlement underscores the importance of timely copyright registration by content creators and adds to the precarious financial picture for enterprise GenAI businesses.

The settlement occurred in a class action lawsuit brought by three authors against Anthropic. The lawsuit alleged that Anthropic used the authors’ books without permission to train Claude and that this constituted copyright infringement. The authors further alleged that Anthropic obtained some of their books for use as training data from two online pirate libraries – Library Genesis and Pirate Library Mirror.

The settlement in the Anthropic case isn’t final. The court has raised concerns about it, but the parties should be able to quickly wrap things up to the judge’s satisfaction.

Also, the settlement addresses only the use of pirate libraries. It leaves open the plaintiffs’ ability to pursue other liability theories. For example, on appeal, they can dispute the court’s earlier ruling that it was fair use (and, thus, not copyright infringement) for Anthropic to use their books in GenAI training where no pirated material was involved.

This case shines a spotlight on popular pirated-content websites.

Here, Anthropic downloaded from Library Genesis and Pirate Library Mirror, which are commonly called LibGen and PiLiMi. In copyright cases against other GenAI makers, another popular pirate site, Z-Library, was also alleged to be a source of GenAI training material.

LibGen and Z-Library have been successfully sued for copyright infringement, but their controllers remain unidentified, and those libraries keep popping back up when taken offline by legal action.

These databases have a wide variety of material copied without permission. LibGen and Z-Library are popular places for college kids to download free electronic textbooks.

As for the settlement, Anthropic agreed to pay $1.5 billion for the first 500,000 books in the pirated libraries it used – but only when a book was copyright-registered within five years of publication and before Anthropic downloaded it. For each book over 500,000, Anthropic agreed to pay an additional $3000 per book.

The authors won’t get that much. The class action plaintiffs’ attorneys are asking for a 25% cut for their fees, and there will be other deductions. I estimate that the authors in the class who meet the copyright registration requirements will get about $2100 per book.

The biggest lesson here is for content creators. If you create anything you don’t want copied without your permission, register your copyright with the U.S. Copyright Office.

You should do this before or immediately after you publish your material (such as your book or photos) to be eligible for “statutory damages.” The availability of statutory damages drove the settlement in the Anthropic case.

If you register your copyright in a work before you publish it, or within 90 days after first publication, or at least before the infringement begins, you could recover statutory damages ranging from $750 to $30,000 per work infringed (such as per book). If the infringement was willful, you could recover up to $150,000 per work infringed.

Also, register your copyright no later than five years after you publish the work. If you don’t do so, the registration will not have as much legal power. Such late registrations are excluded from the settlement pool in the Anthropic case.

What does this settlement mean for businesses?

Stay flexible. Don’t make a long-term contractual commitment to using any particular GenAI, and don’t build business infrastructure that depends on the continued availability of any specific GenAI, such as a chatbot to answer customer service inquiries.

That’s because GenAI providers are in a precarious financial position. They are investing billions in expensive data centers, massive electricity needs, and huge computing costs. They are not yet profitable.

Worse yet, a recent MIT study showed a high failure rate for corporate GenAI projects. While companies have spent about $30-40 billion on AI, 95% of pilot projects didn’t make it past the pilot stage. The study reports that many company-provided GenAI tools do not integrate well with existing workflows and don’t learn or improve over time, leading companies to abandon them.

Relatedly, a U.S. Census Bureau biweekly survey shows AI adoption by companies with more than 250 employees has been dropping since the beginning of the summer.

Still, there’s no doubt that AI will continue to transform work and eventually increase productivity, just as computers and the Internet did. Keep experimenting with it, stay flexible on which one you use, and, authors, register those copyrights!

Written on September 17, 2025

by John B. Farmer

© 2025 Leading-Edge Law Group, PLC. All rights reserved.