In-depth reporting and analytical commentary on artificial intelligence regulation. No legal advice.

Anthropic’s record $1.5B+ copyright settlement is a great deal for book authors that doesn’t sound the death knells on Generative AI

Florian Mueller

Context: Early last week, Claude AI maker Anthropic reached an agreement in principle on a settlement of a book authors’ copyright class action in the Northern District of California (August 26, 2025 ai fray article). The significance of the news was immediately clear, and we took lead class counsel Justin Nelson‘s word for it when he said: “This historic settlement will benefit all class members.”

What’s new: As ordered by Judge William H. Alsup of the United States District Court for the Northern District of California, a settlement motion was filed on Friday (September 5, 2025; PDF). Anthropic will pay a minimum of $1.5B into the settlement fund, and the amount can increase if more than 500,000 copyrighted works subject to the settlement are identified, with a payment of $3K per book. The plaintiffs will not authorize future infringements, and theoretically they could still sue over copyright infringements by Claude AI’s outputs.

Direct impact & wider ramifications: It is in the nature of settlements in a situation like this (months before the trial date and years before any appeals would play out) that the terms reflect not only the plaintiffs’ leverage but also the uncertainty that would have remained if the litigation had continued:

  • For the authors, there was a a risk of winning nothing or only $200 per book (innocent infringement).
  • For Anthropic, the risk was not seriously the $750 billion its lawyers hyperbolically claimed as the product of five million books times $150,000 in maximum statutory damages per work (April 19, 2025 ai fray article), but there was a risk greater than zero of Anthropic losing more than it could afford, at which point the question is whether its investors and partners (such as Amazon and Google) would rather have let the company go bankrupt.
  • Book authors are well advised to take this deal, given that further delay would also entail the risk of legislative intervention by the U.S. government. Even if some authors may believe (and often rightly so) that their books are more valuable than others, the legal basis after Judge Alsup’s summary judgment (June 24, 2025 ai fray article) was such that statutory damages loomed large, but other damage theories were not going to materialize anytime soon (if ever), and it is in the nature of statutory damages that the prohibited act (here, obtaining books through a pirate library) is punished as opposed to a per-title valuation. Bestsellers authors do, however, still have the hypothetical opportunity further down the road of seeking damages based on Claude AI’s output, a theory that was not in this case.

The motion hearing will be held in San Francisco on Monday. Judge Alsup previously encouraged a settlement. He already has senior status and will change this to “inactive” at the end of the year. This largest copyright settlement to date will be part of his legacy without a doubt, and arguably the part that people will remember best and (given how AI is changing the world) for the longest time. Neither side of the AI copyright debate can accuse him of having been biased, given that he sided with Generative AI on the use of copyrighted material for training (which he deemed fair use), but drew the line where outright piracy was involved.

The motion for preliminary approval comes with a proposed schedule. The whole process will extend into next year, and maybe Judge Alsup will actually change mind and volunteer to just wrap up that case. Otherwise, he can easily pass it on to another district judge for the formalities that will remain.

The proposed settlement has been blessed by a former U.S. Copyright Registrar and organizations representing authors.

Lead counsel Justin Nelson issued the following statement:

“This landmark settlement far surpasses any other known copyright recovery. It is the first of its kind in the AI era. It will provide meaningful compensation for each class work and sets a precedent requiring AI companies to pay copyright owners. This settlement sends a powerful message to AI companies and creators alike that taking copyrighted works from these pirate websites is wrong.”

By comparison, the largest copyright verdict so far came down in Oracle v. SAP, but the $1.3B verdict did not stand. In the end, the damages award was but a fraction.

Other Generative AI companies face similar issues, though in Anthropic’s case the connection between the works and the purpose for which they were used (training an LLM on those books “for good writing” used in the automated creation of new books) was particularly strong and clear.

The motion also mentions the herculean effort that the class-action lawyers put into it. This was a hard-fought battle. They went from complaint to principle settlement in a little over a year, but it was no cake walk. In fact, some other copyright cases against AI are struggling because the legal work that was performed on them was nowhere near as good. Anthropic, with backing from Amazon and Google, hired an army or armies of lawyers, including some extremely well-known ones, trying to turn this into a battle of materiel. Class counsel needed enormous resources to wage this battle and secure a great outcome.

AI providers will defend against some cases, sometimes just because of legal deficiencies. Some cases will run much longer, and then there is a possibility of appellate clarification of key questions, particularly relating to fair use. The Bartz v. Anthropic settlement upholds the fundamental values of copyright without jeopardizing the viability of Generative AI and impeding technological progress.

Class counsel

The following lawyers were named in the motion for preliminary approval of the settlement and/or the motion for class cetification, and for sure many more were contributing but not named in court documents: