On June 23, 2025, Judge Alsup issued a much-anticipated summary judgment ruling in a copyright case brought by a group of authors seeking to challenge Anthropic’s alleged copying of millions of books in connection with Anthropic’s development of a generative artificial intelligence (“GenAI”) model called “Claude.” Bartz v. Anthropic PBC, No. 3:24‑cv‑05417‑WHA (N.D. Cal. June 23, 2025).

In the Court’s 32 page Order, the court declared GenAI models such as Claude “among the most transformative [technologies] many of us will see in our lifetimes,” and found that using lawfully acquired books to tokenize them and use the resulting tokens to develop a GenAI model is fair use. The court’s decision was based on findings that Claude does not reproduce plaintiffs’ works as output nor does it displace demand for the originals. The court observed that while a market may develop for licensing works for machine learning purposes “such a market … is not one the Copyright Act entitles Authors to exploit.” Order at 28. 

In reaching its conclusion, the court relied in part on Authors Guild, Inc. v. HathiTrust, 755 F.3d 87, 97, 101, 103 (2d Cir. 2014). In that case, the Second Circuit affirmed summary judgment that mass digitization of millions of books to create a full‑text searchable database and to provide accessible copies for print‑disabled users was a transformative fair use.¹ 

Judge Alsup's decision was not entirely favorable to Anthropic, however. The court denied summary judgment to Anthropic on its downloading and maintenance of a database of pirated books (i.e., a dataset of books that was copied without any payment of any kind to the copyright holders). The court found issues of fact and noted that it will send issues relating to the pirated dataset to trial.

The decision represents a notable departure from the last significant ruling in this space, namely Thomson Reuters Enter. Ctr. GMBH v. Ross Intel. Inc., 765 F. Supp. 3d 382 (D. Del. 2025), motion to certify appeal granted, No. 1:20-cv-613-SB, 2025 WL 1488015 (D. Del. May 23, 2025). In Ross, plaintiff Thomson Reuters—the company behind the Westlaw legal research database—argued that the defendant illegally trained a competing AI legal search tool on source material pulled from Westlaw’s proprietary headnotes. 

The court in Ross ultimately granted summary judgment for Thomson Reuters after rejecting Ross’s fair use defense. It did so largely based on a conclusion that use of plaintiff’s headnotes to create a competing legal search tool was not transformative. 

Judge Alsup neatly distinguished Ross based on a finding that Claude is far more transformative than the tool at issue in Ross. In Ross, what was trained was merely “a competing AI tool for finding court opinions in response to a given legal topic.” Order at 13. In contrast, Claude can be used to generate new text in a manner that is “quintessentially transformative.” Id.

Judge Alsup also takes a very different approach on market harm than that adopted by the court in Ross. Specifically, in Ross the court appears to endorse a market dilution theory, i.e., the idea that widespread, uncompensated use of copyrighted material can erode the value of actual or potential licensing markets by undermining the expectation of payment. 765 F. Supp. 3d at 400.

Judge Alsup implicitly disagrees, finding that since Claude does not output verbatim versions of plaintiffs’ works it “did not and will not displace demand” for plaintiffs’ books specifically. Judge Alsup noted that while the output generated by Claude may in fact compete with plaintiffs’ works that is not an injury within the purview of the Copyright Act. Order at 28.

In sum, Judge Alsup’s decision takes a very different approach from that adopted by the court in Ross and presages the unpredictable road ahead for the myriad cases across the country at the crossroads of fair use and AI.

Footnotes