Context: Last summer, book authors brought a class action against Claude maker Anthropic (August 20, 2024 ai fray article). Claude is known for its ability to generate particularly large documents, such as long texts.
What’s new: There have recently been some filings in this case (Bartz v. Anthropic) that show the litigation is heating up at an early stage where other lawsuits tend to be rather uneventful. Anthropic is delaying the production of documents that are, however, considered necessary for the class-action plaintiffs to respond to Anthropic’s summary judgment motion. Anthropic tries to get the entire case thrown out at this early stage on the basis of the fair use exception to copyright law. When Judge William H. Alsup of the United States District Court for the Northern District of California saw that Anthropic had once again failed to keep a procedural promise, he entered a scheduling order that shortens Anthropic’s time to write a reply brief to four days. Separately, Judge Alsup raised some questions about the (il)legality of using pirated material for an allegedly transformative purpose. Separately, another major development in AI copyright litigation is the consolidation of all cases against OpenAI in the Southern District of New York.
Direct impact: Anthropic was even at risk of its summary judgment motion being denied. At this point, Anthropic still has the chance that its motion will be adjudicated, but it has to get its discovery act together. The questions Judge Alsup has raised about the impact of piracy on fair use are nonjudgmental. The class-action plaintiffs, however, point to case law that says there cannot be fair use of pirated material. Separately, the decision that all OpenAI cases are put before a judge in the Southern District of New York does not make a particular outcome more or less likely. In fact, some plaintiffs preferred that district anyway, among them the New York Times.
Wider ramifications: In a different case in a different district (Delaware), ThomsonReuters, the owner of the Westlaw legal research service, defeated a summary judgment motion by an AI provider (Ross in that case) centered around fair use (February 11, 2025 ai fray article).
Instead of annoying Judge Alsup, Anthropic should actually be grateful that he agreed to rule on Anthropic’s fair use motion, which is all about fair use, before class certification. But he expected Anthropic to meet its discovery obligations, which it apparently hasn’t. And it’s not even the first such failure: Anthropic originally promised to meet its obligations by March 20, 2025.
This is about documents that the class-action plaintiffs’ counsel needs to see before responding to Anthropic’s summary judgment motion. It is presumably very relevant to the question of whether Anthropic breached copyright law through how it obtained the material it used to train its large language model. Summary judgment does involve factual questions, not just legal theories. Decisions on motions to dismiss can also have factual elements, but they are typically more about the pleadings, with factual allegations merely having to be plausible. For the plaintiff side to effectively oppose the summary judgment motion, access to some fundamental facts is key.
After Anthropic first broke its promise (the one to put the requested material on the table by March 20), the court held an additional conference. The plan was then for Anthropic to finally deliver what it promised to provide, and the plaintiffs were given an extension of time to file their opposition to the summary judgment motion. But yesterday the plaintiffs wrote a letter to Judge Alsup (PDF) because Anthropic still hadn’t met its obligations. Apparently Judge Alsup thought that enough was enough. Without even awaiting (much less inviting) a response by Anthropic, he entered an order (PDF) that took the AI company to task over its behavior. The order culminated in a rebuke of Anthropic’s counsel’s conduct:
“When lawyers make agreements they must honor their word.”
That does not mean to suggest an impact on the chances of Anthropic’s motion. Judge Alsup will separate his personal disappointment in this behavior from the merits of the motion. But Anthropic has at least one problem regardless: its delay game does not excude confidence in its position that it delays the production of documents and also appears to make excessive use of the instrument of attorney-client privilege assertions. Anthropic comes across as a party that has something to hide. Apparently Anthropic wants to impair the plaintiffs’ ability to review the material and use it to underpin its arguments against the Anthropic’s fair use theory. That just doesn’t look good.
Anthropic responded to the order by seeking reconsideration. Judge Alsup will hold a hearing on Wednesday (April 9) to discuss the matter with counsel.
Judge Alsup already made a clear distinction between the merits of the case and the helpfulness of the lawyers involved. At an October 10, 2025 hearing (the transcript of which has recently become available for download on PACER), he told the following to Susman Godfrey’s Justin Nelson, the Bartz plaintiffs’ lead counsel:
A very good, short summary. You get an A plus. I don’t have to agree with everything, but you did what I asked. In two minutes or less, you summarized the case.
Mr. Nelson received even greater praise in the famous Dominion v. Fox News case, which ended in a $800M settlement and where the judge said this was the best lawyering he had ever seen (April 18, 2023 New Yorker article). But Judge Alsup is one of the most difficult judges to please. And Anthropic is presently on the opposite end of the spectrum from pleasing the court.
Another important order that came down yesterday (PDF) was not about anyone’s conduct, but all about the substance of the case. It addressed the single question that could prove dispositive of Anthropic’s fair use defense, which is whether someone who illegally obtained copyright material can later argue that they made transformative use of it and, therefore, are in the clear:
Somewhere in the briefing on summary judgment each side shall please explain the extent to which each of the following cases does or does not violate the copyright laws:
Case 1. An e-book is purchased at full price and read over five days by the purchaser. Each day he reloads the entire e-book.
Case 2. Same as Case 1 but he buys the e-book for one cent from a notorious counterfeiter known as Books ‘R’ Cheap.
Case 3. Same as Case 2 but his purpose throughout is to write a transformative parody of the work (and he does).
Case 4. Same as Case 3 but his purpose further includes to write the parody with cowriters, so he buys one e-book and copies it for all co-writers (they jointly write the parody).
Case 5. Someone buys a copy of the e-book at full price and then gives it to a writer who uses it to write a transformative parody of the work.
Variant. For any case relevant (or for all at once), briefly explain to what extent it makes any difference for the copyright laws whether the work is purchased as an e-book or instead in print and converted by the purchaser immediately into a digital format before the next step.
Those fictitious scenarios may be useful just to understand where U.S. copyright law stands. Anthropic’s lawyers point to decisions in which, for example, copying a game console operating system without authorization was allowed in order to enable a game developer to make third-party games. But that is just Ninth Circuit (regional) case law. In the end this question will have to be addressed for AI systems by none other than the Supreme Court.
In their motion for class certification (PDF), the plaintiffs distinguish between two clasess: a “Pirated Books Class” (authors of books that Anthropic obtained through a collection of pirated material) and a second class, the name and definition of which is so heavily redacted that it is not possible to comment on that one for the time being. In connection with pirated books, the plaintiffs say they are “aware of no case — in any jurisdiction at any time — where a court found that the initial illegal download [emphasis in original] was fair use due tot he use to which the illegal downloaded material was subsequently put.” In that context, they point to a decision by the famous Judge Richard Posner who upheld a preliminary injunction in a 2003 Seventh Circuit case (In re Aimster Copyright Litigation).
It will be interesting to see how Judge Alsup will resolve the fair use question on summary judgment. Whatever he decides, there will be an appeal. If he granted summary judgment and threw out the case, the matter would go straight to the Ninth Circuit and, potentially, from there to the Supreme Court.
All OpenAI cases go to New York
Not only in Bartz v. Anthropic but also with respect to a dozen OpenAI copyright cases was it an eventful day. A multidistrict litigation panel decided (PDF) that all of a dozen AI copyright cases brought against OpenAI should be put before Judge Sidney H. Stein of the United States District Court for the Southern District of New York. The highest-profile OpenAI case that has been filed there so far is the one that the New York Times brought in 2023. There is already a book authors’ case (that resulted from the merger of two cases, one by fiction and one by non-fiction authors) in that district: Authors Guild v. OpenAI. Mr. Nelson is also lead counsel there. For him, nothing changes. Some others have to commute from California to New York now. That means some coast-to-coast flights to hearings and, ultimately and potentially, trials.