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

$1.5B Anthropic-book author settlement: approval postponed, should never be taken for granted, but trial still appears unlikely

Context: Last month, Claude AI maker Anthropic and book authors announced an agreement in principle on a settlement of a copyright class action in the Northern District of California (August 26, 2025 ai fray article). The terms, which involve a minimum payment of $1.5 billion ($3,000 per book) that looks like pretty good compensation for book authors but is far from threatening the viability of Generative AI, became known at the end of last week (September 6, 2025 ai fray article).

What’s new: Judge William H. Alsup of the United States District Court for the Northern District of California does not consider the initial settlement proposal ripe for approval and has given the parties some homework with deadlines on the next two Mondays: September 15 for the class list and works list, and September 22 for other clarifications such as regarding the claim form and claims procedure. The next hearing will be held on September 25. After the hearing, the Authors Guild and the Association of American Publishers (AAP) issued statements criticizing the court.

Direct impact: While Judge Alsup reportedly announced a “denial without prejudice” at the hearing, the post-hearing minute entry says “[p]reliminary approval was postponed pending submission of further clarifying information.” Therefore, it is reasonable to assume that the case will be settled. Judge Alsup warned the parties that unless they satisfy his requirements over the next couple of weeks, the case will proceed to a December trial as planned. But statistically that is very unlikely, and a combination of more specifics and further clarifications and explanations should normally result in an approved settlement. Judge Alsup already suggested in the spring that the parties should settle (May 23, 2025 ai fray article), and warned Anthropic against potentially “los[ing] big” (August 11, 2025 ai fray article).

Wider ramifications: Federal judges have been criticized for rubberstamping too many settlement proposals, even in that particular district (e.g., April 18, 2003 citizen.org article). But in recent years, the bar has been considerably higher (March 15, 2019 Los Angeles Times article). It is not surprising that a record settlement such as this one would undergo judicial scrutiny, but there are objective reasons for which a settlement along the lines of the Anthropic agreement is in the public and the parties’ interest.

Based on what the various reports say, Judge Alsup is a neutral gatekeeper. A strict one, but impartial. He wants to shield Anthropic from the risk of an unforeseeable level of claims by “hangers on” while also ensuring that book authors entering into this settlement know what they are doing. None of that can be reasonably criticized.

Post-hearing statements by the Authors Guild and the AAP must be attributed to those organizations’ frustration over having been accused of striking a deal on authors’ behalf without sufficiently involving the actual stakeholders. They obviously don’t want anyone or anything to drive a wedge between themselves and the right holders they represent.

The one who really stands to lose the most if things were to surprisingly go wrong is still Anthropic. For book authors, it’s a very good deal. For class-action lawyers, it’s about a historic success they’ve worked hard to achieve. It would be unfortunate for the plaintiff side if the settlement didn’t materialize, but their chances of achieving a similar outcome at trial are not bad, though it would then take years of delay and protracted uncertainty with appeals and hypothetical retrials before anyone would get a cent. What’s at stake for Anthropic, however, is the risk of a devastating verdict. The company’s valuation is high and its backers (Amazon and Google) have deep pockets. But there comes a point where the financial implications would have Anthropic’s investors and partners worried.

Therefore, Judge Alsup’s warning that the case might go to trial unless the settlement is good enough to be approved puts pressure on everyone, but Anthropic with its vast resources now has a particularly strong incentive to ensure that the September 15 and 22 deadlines are met and that the data and clarifications provided to the court will lead to an approval.

Publishers are right that a settlement should put a matter to rest rather than spawn new lawsuits. But reason will presumably prevail. If they can convince Judge Alsup that a certain structure would give rise to a wave of litigation against publishers and authors of jointly-owned works, and between co-authors, then he will enable a settlement on workable terms. However, if that risk is merely hypothetical and in practice there are already agreements in place that can be applied to the revenue distribution resulting from the settlement, then the judge will probably not be dissuaded from insisting on certain terms and clarifications.

It’s hard to see a problem of authors not hearing about this and knowing there’s money waiting them. Class notification must be well-structured, but the copyright holder community is extremely aware of this. There has been a lot of media coverage already, and when there is an approval of a settlement and when payouts begin, there will be even more media coverage.

As far as legal fees are concerned, it’s interesting that Judge Alsup on the one hand wanted additional people to participate in yesterday’s hearing, but on the other hand says the fees for certain lawyers who came on board in connection with the settlement would not be paid out of the settlement fund. A solution will be found for that. And as far as the legal fees in general are concerned, class counsel took enormous risks. Arguably, what Judge Alsup said about Anthropic would lose big if its wrongdoing was big also applies to that. If those lawyers picked up a high-value cause, it’s because of the scope and scale of the wrongdoing that led to this litigation in the first place.

There is no shortage of book authors in the world. The author of this article has written about a dozen books, too, though they are not at issue in that case. So there will always be some authors who publicly criticize such a settlement. The way to look at it is that it’s easier to demand more than to achieve more.

Whether one agrees with it or not, and no matter how one thinks the United States Court of Appeals for the Ninth Circuit or potentially the Supreme Court of the United States would adjudicate the “fair use” issue in the case, it is a practical reality that Judge Alsup’s summary judgment (June 24, 2025 ai fray article) makes this a statutory damages matter. Everything else would be two birds in the bush as opposed to one bird in the hand, and that even looks like an understatement. From a probabilistic point of view, Anthropic had a slim chance of avoiding a near-term trial, at which it would then have been very difficult for Anthropic to get away with a small amount per book (such as based on innocent infringement), but there were residual risks for the book authors and, as explained before, getting compensated for the use of their works in LLM training was impossible without first getting Judge Alsup’s summary judgment overturned.

If some authors would prefer the kind of deal that reflects a traditional copyright license agreement, these are their options:

  1. They might think about it in practical terms and take the payout based on a realistic assessment of how difficult and costly it would be, and what delays it would entail even in a positive scenario, if they were to litigate. Furthermore, they might take comfort in the fact that infringements by output are not in this case and not part of the settlement, so depending on how the case law evolves, there could be an opportunity for incremental income.
  2. They can stay out of this deal and try their luck.

The standard is not and cannot be that the settlement is only a good one if everyone agrees. But if it is widely accepted, that will be in Anthropic’s interest.

The settlement will definitely have to reflect Judge Alsup’s guidance. But if it were to fall through, it would be difficult to find a jury, especially in the tech-savvy Northern District of California, where no one would be aware of what happened. That is another practical consideration for which this should work out, and presumably will.

Counsel and other hearing participants

Plaintiffs’ Attorneys:

Defendant Attorneys:

Class Plaintiff: Andrea Bartz, Charles Graeber, Kirk W. Johnson.

Authors’ Coordination Counsel: Cowan, DeBaets, Abrahams & Sheppard’s Nancy Wolff and Fairmark’s Yinka Onayemi.

Publishers’ Coordination Counsel: Edelson PC’s J. Eli Wade-Scott and Oppenheim + Zebrak’s Matt Oppenheim.

Authors Guild CEO: Mary Rasenberger.

Association of American Publishers CEO: Maria Pallante (former U.S copyright chief).