The Free Software Foundation (FSF) has received notification regarding a proposed settlement in the class action lawsuit Bartz v. Anthropic, a case centered on copyright infringement allegations stemming from the use of copyrighted works to train large language models (LLMs). The lawsuit claims Anthropic unlawfully downloaded books from online repositories like Library Genesis and Pirate Library Mirror to develop its AI technologies.
According to the FSF, the district court previously ruled that Anthropic’s use of the books to train its LLMs constituted “fair use” under copyright law. However, the legality of downloading the books in the first place was left unresolved and slated for trial. Rather than proceed to trial, Anthropic is now initiating contact with copyright holders, including the FSF, to offer financial compensation in lieu of potential damages.
The FSF holds copyrights to numerous programs within the GNU Project, as well as several books, all published under free licenses that permit use without payment. One such work, “Free as in freedom: Richard Stallman’s crusade for free software,” co-authored by Sam Williams and Richard Stallman, was identified as being present in the datasets used by Anthropic to train its LLMs. The book is licensed under the GNU Free Documentation License (GNU FDL).
The FSF expressed concern over the broader implications of LLM training practices, advocating for the sharing of complete training data, models, configurations, and source code with all users. In a statement, the organization urged Anthropic and other LLM developers to release their models “in freedom,” ensuring users have access to the underlying components and data used in their creation.
The case, initially filed in August 2024 by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, alleged that Anthropic relied on illegally obtained e-books – sourced from so-called “shadow libraries” – to train its Claude LLMs. Judge William Alsup’s June 2025 summary judgment ruling distinguished between the fair use of copyrighted material for training purposes and the infringement associated with acquiring pirated copies. The ruling found that Anthropic’s digitization of lawfully purchased books for its central library was similarly a fair use.
Despite the fair use finding for training, Judge Alsup expressed skepticism about justifying the initial downloading of pirated materials, suggesting that such actions could undermine any subsequent fair use defense. This nuance positioned the case for a potentially costly trial, as Judge Alsup certified a class of U.S. Copyright holders potentially numbering in the millions, with statutory damages potentially reaching $150,000 per infringed work. Anthropic’s attempts to appeal the decision and secure a stay were denied, leaving a December trial date looming.
The FSF stated that, while a small organization with limited resources, it would seek user freedom as compensation should it participate in a similar lawsuit and identify its copyrights or licenses violated. The organization has not yet indicated whether it will formally participate in the settlement process.
Leave a Reply