On June 23, 2025, in Bartz v. Anthropic PBC, C 24-05417 WHA (N.D.Ca.), the United States District Court for the Northern District of California (the Court) held that Anthropic PBC’s (Anthropic’s) use of legally obtained works to train its artificial intelligence (AI) models was “transformative” and constituted fair use under the Copyright Act (the Act). However, this did not extend to Anthropic’s use of pirated copies for such training, which was not protected under fair use.
Anthropic is a software firm that offers an AI service, Claude, which was trained on millions of books and other texts selected and assembled into a “central library” by Anthropic. The central library materials were sourced through bulk-purchasing print books that were scanned and subsequently discarded; and obtaining unauthorized “pirated” copies of certain books. Andrea Bartz et al. (collectively, the Plaintiffs) are authors of some of the books in Anthropic’s central library and brought a class action against Anthropic alleging copyright infringement. Prior to the Court certifying the class action, Anthropic moved for summary judgement on whether its actions qualified as fair use and were protected under the Act.
The Court found that the use of legally obtained works to train specific AI models constituted fair use. This was largely due to the training of AI models being a transformative process with the Court acknowledging that the “technology at issue was among the most transformative many of us will see in our lifetimes.” However, the Court found that Anthropic’s sourcing of pirated copies for its central library was not fair use, especially when certain pirated materials were retained by Anthropic and not used for AI training. Consequently, the Court granted summary judgement in favor of Anthropic’s AI training using legally obtained works as fair use but denied summary judgment for Anthropic with respect to the pirated library copies.
This decision may signal the start of a broader trend of AI companies successfully relying on fair use protection when training AI models with copyrighted works. A few days following this decision, another case, Kadrey v. Meta, Case No. 23-cv-03417-VC, was released where it was held that Meta’s use of certain copyrighted works for AI training was protected by fair use.
Summary By: Uday Bahal
Disclaimer: This Newsletter is intended to provide readers with general information on legal developments in the areas of e-commerce, information technology and intellectual property. It is not intended to be a complete statement of the law, nor is it intended to provide legal advice. No person should act or rely upon the information contained in this newsletter without seeking legal advice.
E-TIPS is a registered trade-mark of Deeth Williams Wall LLP.