Context: Judge William H. Alsup of the United States District Court for the Northern District of California granted Anthropic’s motion for summary judgment against a book authors’ class action with respect to LLM training, but denied it with respect to the use of pirated material (June 24, 2025 ai fray article). He also certified the class, letting the case go forward. The hypothetical amount of statutory damages that Anthropic might owe could be on the order of a trillion U.S. dollars (April 19, 2025 ai fray article). The Google- and Amazon-backed AI provider is scared to death and wants to at least slow things down by means of a petition for an interlocutory appeal (July 15, 2025 ai fray article), which certain technology lobbying groups are supporting (August 9, 2025 LinkedIn post by ai fray).
What’s new: Judge Alsup has denied Anthropic’s motion to stay the district court proceedings pending the resolution of the defendant’s appellate ambitions. Besides explaining his take on why an interlocutory appeal is not going to help Anthropic, Judge Alsup also rejects Anthropic’s suggestion that he wanted to “rush” the case only because he plans to switch from senior to inactive status later this year.
Direct impact: Judge Alsup’s order is major setback for Anthropic in various ways, even though the denial of the motion was probably the expected outcome. Judge Alsup’s decision provides ammunition for the plaintiffs as Anthropic has made, or will soon make, the same or similar arguments in the appeals court (Ninth Circuit). With a view to the further proceedings, it sheds light on the state of affairs, suggesting that Anthropic indeed faces considerable risk of a devastating damages award (which it could only avoid by settling now). And one does not even have to read between the lines to see that Judge Alsup is annoyed by Anthropic’s disrespectful attitude, though the latter may be borne out of despair.
Wider ramifications: Technology industry lobbyists and their lawyers are arguing that a settlement under which Anthropic would pay a substantial amount of money to the class-action plaintiffs could adversely affect AI innovation. But that is hyperbole. In the end, there is enough money flowing into AI that Anthropic would survive the settlement and continue to be a viable market actor. The only organizations that have anything to fear are those who, like Anthropic, used pirated material.
Here’s Judge Alsup’s order, which is arguably no less significant than his mixed ruling on fair use, followed by some observations and explanations:
There are two groups of judges when it comes to what an appeals court will make of their decisions. There are those who just decide what they think is right and don’T care if they get overruled. Judge Lucy H. Koh had that attitude when sitting on the United States District Court for the Northern District of California, and is now a Ninth Circuit judge. Judge Alsup, also of the Northern District of California (though the San Francisco, not San Jose division) is known to be on the other end of the spectrum. He very much does care, and has shown it in various situations. If there was any doubt about it, the following sentence in the order he handed down this morning (quite early on the West Coast, but he’s known to be an early riser) would eliminate it:
To assist the court of appeals, this order now explains why a stay
would be ill-advised.
He then dismantled Anthropic’s arguments. The first section, which has multiple subjections, deals with Anthropic’s claim that it is likely to succeed on the merits.
Judge Alsup does not see that Anthropic’s download and storing of large volumes of pirated material can be excused with fair use, at least not without a fully developed record:
The machinations of Anthropic’s downloading of pirate libraries and its deployment of bit-torrenting to do so looms large in this case and should be fully before the jury, district judge, and court of appeals before we say the definitive extent to which this copying of pirated materials was fair or not.
Here’s why Anthropic may actually be unable to establish fair use in the piracy context:
Anthropic recopied only some “subsets” of the pirated works to use for training large language models, or LLMs, but kept them all. Which works in its collection were actually used to train LLMs, which were not, and why were all retained? Anthropic has refused to come clean on this, even now, and for all we know, most were never used (or not solely used) to train LLMs.
This looks like a typical case of “if the facts are not on your side, argue the law.” Anthropic may know that it will be in terrible shape if this case proceeds to trial, and it may indeed be unable to take that risk. It may have no choice but to settle. So it needs a broadbrush, bright-line legal standard to be let off the hook:
Anthropic asks our court of appeals to ignore all these colorations in its piracy. It seeks a sweeping rule, as a matter of law, that Anthropic was entitled to pirate all the copyrighted works it wanted and to keep them indefinitely so long as any part of the trove was further copied and used to train an LLM. This proposed rule cannot be reconciled with the third statutory fair use factor, the amount and substantiality of the portion taken for the claimed purpose.
After pointing Anthropic to its own representations at the summary-judgment stage, Judge Alsup effectively accuses the company of stonewalling:
To this day, Anthropic has been unable or unwilling to say which of the pirated works were actually used to train LLMs. It is entirely possible, to repeat, that only a small subset of the pirated works was ever so used.
Thereafter, Judge Alsup clarifies that, contrary to Anthropic’s misrepresentations, he has not prejudged the case and Anthropic can still argue fair use at trial, but he apparently (and understandably) believes Anthropic doesn’t want to come clean.
As for Anthropic’s attempt to get the class certification overruled by the Ninth Circuit, Judge Alsup doesn’t understand why a stay of the proceedings in his court should even be needed. Even the timelines Anthropic itself provided would enable the appeals court to decide ahead of trial. But what he can’t accept is that Anthropic attributes his decisions on the legal standard to ulterior motives:
Anthropic suggests that the district judge is “rush[ing]” this case so he can retire. It is true that the district judge would like to take inactive status near the end of this year or shortly thereafter. But it is not true that the district judge is rushing this case. The trial date was set long ago at the initial case management conference and has not changed since. Keeping a case on track is not the same as rushing the case to trial.
Anthropic wouldn’t antagonize the district judge if it wasn’t looking at the combination of a potentially massive damages award and the fact that if it somehow avoids the December trial, Judge Alsup will have retired fully, so Anthropic won’t have to deal with him anymore and it doesn’t matter whether he’s upset. Unfortunately for Anthropic, there is some solidarity among and between judges, so what Anthropic has already said and done may be viewed very negatively be the appeals court and by whoever will handle the case should there be a delay or (after an appeal) a remand.
Judge Alsup isn’t buying Anthropic’s calculation according to which it would not financially survive a damages verdict:
Anthropic’s doomsday calculation is premised on every eventuality coming out completely against its druthers. The crux is its assumption that Anthropic will lose and lose badly at every trench, and that this will lead it to have to post a bond for appeal so large as to kill the entire company. This piles assumption upon assumption. For all we know at this stage, Anthropic will persuade the jury to find facts vindicating it completely (CC Order 28). But if Anthropic loses big it will be because what it did wrong was also big.
The last sentence is huge, which is why it is quoted in the headline of this article. The penultimate sentence must be interpreted correctly: this does not mean Judge Alsup thinks the facts weigh in Anthropic’s favor, but that Anthropic intends to argue to the jury that the facts vindicate it completely. Between the lines, he’s saying that Anthropic doesn’t appear to believe in its own defenses as much as it otherwise claims.
Obviously, the next step is that Anthropic will now ask the appeals court for a stay, and Judge Alsup’s order ups the ante for Anthropic.
Counsel
Counsel for Anthropic
- Cooley’s Kathleen Hartnett, Ephraim McDowell and Alexander J. (“Alex”) Kasner.
- Arnold & Porter Kaye Scholer, San Francisco: Douglas A. Winthrop, Joseph Farris, Jessica L. Gillotte and Estayvaine Bragg.
- Arnold & Porter Kaye Scholer, Los Angeles: Angel T. Nakamura, Oscar Ramallo and Allyson Myers.
- Stanford professor Mark Lemley of Lex Lumina, who has been working for Google (a key Anthropic partner) for decades.
- Latham & Watkins, San Francisco: Joseph R. Wetzel and Andrew M. Gass.
Counsel for the book authors
- Susman Godfrey’s Justin A. Nelson (lead counsel; he’s involved with virtually every major AI copyright case, on the plaintiff side, apart from certain cases against Meta), Rohit Nath, Alejandra C. Salinas, Collin Fredricks, Jordan W. Connors and J. Craig Smyser.
- Lieff Cabraser Heimann & Bernstein’s Rachel Geman (co-lead counsel), Jacob S. Miller, Danna Z. Elmasry, Daniel M. Hutchinson and Reilly T. Stoler.
- Cowan DeBaets Abrahams & Sheppard’s Scott J. Sholder and CeCe M. Cole.