FSF on Anthropic Copyright Settlement & LLM Training Data
The Free Software Foundation (FSF) rarely pursues legal action regarding copyright, but when it does, the organization prioritizes securing freedoms for users – even in settlements. This approach is playing out in the ongoing class action lawsuit, Bartz v. Anthropic, concerning the apply of copyrighted materials to train large language models (LLMs). The FSF recently received notice regarding a settlement in the case and, while not directly involved in the litigation, is positioned to benefit as a copyright holder.
The lawsuit centers on Anthropic’s practice of downloading books from online repositories like Library Genesis and Pirate Library Mirror to train its LLMs. While a district court ruled that using the books for training constituted “fair use” under US copyright law, the legality of downloading the books in the first place remained unresolved. Rather than proceed to trial on that point, Anthropic has opted for a settlement, offering financial compensation to copyright holders.
The FSF’s Stake in User Freedom
The FSF holds copyrights to numerous programs within the GNU Project, as well as several books, including Sam Williams’s Free as in Freedom: Richard Stallman’s Crusade for Free Software. Crucially, the FSF publishes all its copyrighted works under free licenses – licenses that permit use, modification and distribution without payment. Williams’s book, for example, is published under the GNU Free Documentation License (GNU FDL).
The FSF’s position isn’t simply about financial compensation. The organization believes the ideal outcome would be for LLM developers like Anthropic to share the complete training data – including the source code, model configurations, and datasets – with users, ensuring “freedom” in the sense of control and understanding. As the FSF stated in a recent blog post, “the right thing to do is protect computing freedom.”
Understanding the Broader Bartz v. Anthropic Settlement
The Bartz v. Anthropic case has develop into a landmark event in the rapidly evolving intersection of copyright law and artificial intelligence. Initially filed in August 2024 by three authors – Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson – the lawsuit quickly expanded to encompass a class of nearly half a million copyright holders, according to reporting from Wolters Kluwer Legal & Regulatory. The initial claim alleged that Anthropic had illegally downloaded millions of copyrighted books from “shadow libraries” – websites offering unauthorized access to copyrighted material – to train its Claude language models.
The court’s split decision in June 2025, distinguishing between the legality of using acquired books for training versus illegally downloading them, significantly shaped the settlement negotiations. Judge William Alsup found that utilizing legally obtained books for AI training qualified as “transformative” fair use, a key legal concept allowing limited use of copyrighted material without permission. However, downloading pirated copies was deemed a clear infringement.
In August 2025, Anthropic agreed to a settlement valued at $1.5 billion, making it the largest copyright settlement in U.S. History, as reported by Wolters Kluwer. The settlement fund is now being distributed to eligible copyright holders, with a significant portion allocated to legal fees. The Authors Guild is actively working to gather information on books that were rejected due to publishers’ failure to register copyright, as detailed on their website.
The Implications for Free Licensing
The FSF’s involvement highlights a critical tension: the clash between traditional copyright models and the principles of free software licensing. The GNU FDL, under which Free as in Freedom is published, explicitly permits use for any purpose, including training AI models, without requiring permission or payment. This contrasts sharply with the standard copyright restrictions that formed the basis of the Bartz v. Anthropic lawsuit.
The FSF’s stance suggests that simply providing financial compensation isn’t sufficient. True resolution, in their view, requires a commitment to transparency and user empowerment – sharing the tools and knowledge necessary to understand and control the technology being developed. This aligns with the broader free software movement’s emphasis on user freedom and collaborative development.
Fair Use and the Future of LLM Training
The court’s ruling on fair use in Bartz v. Anthropic sets a precedent for how copyrighted material can be used in the development of AI models. While downloading pirated content remains illegal, the use of legally acquired books for training is now considered protected under fair use doctrine. This distinction is crucial for LLM developers, who rely on vast datasets to improve the performance of their models.
However, the legal landscape remains complex. The ruling doesn’t address all potential copyright concerns related to LLM training, such as the use of copyrighted code or the potential for models to reproduce copyrighted content. Further litigation is likely as AI technology continues to evolve.
The FSF acknowledges its limited resources and recognizes it must prioritize its legal battles. However, the organization has made clear that it will actively defend its copyrights and licenses, and advocate for user freedom, should its works be used in ways that violate its principles. The outcome of the Bartz v. Anthropic settlement, and future legal challenges, will undoubtedly shape the future of copyright and AI.
Looking Ahead: The settlement process is ongoing, and copyright holders are currently submitting claims. The long-term impact of the Bartz v. Anthropic case will depend on how courts interpret the fair use doctrine in future AI-related litigation and how LLM developers respond to the growing demand for transparency and user control.