FSF Challenges Anthropic LLM Copyright: The Demand for User Freedom

Anthropic recently settled the Bartz v. Anthropic lawsuit, which addressed claims of alleged copyright infringement during Anthropic LLM training. The Free Software Foundation (FSF) was informed of potential financial entitlement from this settlement, as its copyrighted book, "Free as in Freedom: Richard Stallman's Crusade for Free Software," was found in Anthropic's training data. The FSF, however, was not directly involved in the lawsuit.

In the Bartz v. Anthropic case, the legal debate often centered on "fair use," with the district court ruling that training LLMs with books qualified as fair use. This ruling, however, has a key nuance. While the act of training might be fair use, the initial downloading and storage of copyrighted material for that training could still be infringement. This distinction is important because it separates the transformative use of data from how that data was legally acquired, especially concerning Anthropic LLM development.

The FSF's Demand for "User Freedom" in Anthropic LLM Training

The FSF's stance, recently announced, moves away from the idea of monetary compensation. The book, "Free as in Freedom," was published under the GNU Free Documentation License (GNU FDL). This is a copyleft license. It allows free duplication and distribution, but it also demands that any derivative works or modified versions must be released under the same license. Copyleft acts like a viral license: build on something licensed this way, and your new creation must offer the same freedoms.

If the FSF were directly suing for copyright infringement of its copyleft material, it wouldn't seek money. Instead, the FSF would ask for "user freedom." This means the infringing party—like Anthropic—would need to release several things under a free license: the complete training inputs, the full Anthropic LLM model, its training configuration settings, and all accompanying software source code.

The FSF's position is direct: "we don't usually sue for copyright infringement, but when we do, we settle for freedom." This approach directly challenges the proprietary nature of current LLM development, where companies like Anthropic rarely disclose their training datasets or model architectures.

Industry Perspectives and the Road Ahead

The FSF's position is notable, particularly as the FSF was not directly involved in the Bartz v. Anthropic lawsuit. This leads to its statement being viewed as a hypothetical position rather than an immediate legal threat, especially with financial settlements dominating the news.

This sentiment points to a wider industry tension. For many AI companies, a significant financial settlement might just be a "cost of doing business." It's a significant but manageable expense to keep proprietary control over their valuable models. The FSF's demand, however, fundamentally challenges this business model. If an Anthropic LLM is considered a "derivative work" of copyleft-licensed material, then licenses like the GNU FDL or GNU General Public License (GPL) would apply. This would mean models like Claude, their training data, and code would need to become openly available—a direct contradiction to the closed-source approach many commercial AI developers prefer.

Beyond legalities, practical questions arise for AI development. Content creators worry that AI companies are ignoring existing copyright frameworks. These companies often defend their practices by claiming they "only scraped links" or that "AI didn't reproduce exactly 100% of the content" to downplay infringement.

What to Watch

The FSF's position, while not a direct legal action in the Bartz case, illuminates a fundamental philosophical and practical divide within the AI industry. It forces a critical examination of Anthropic LLM intellectual property status: are they truly transformative new creations, or are they derivative works bound by the licenses of their training data? This question extends beyond legal definitions, challenging our understanding of creation itself in the age of AI.

For developers and companies, this evolving landscape demands close attention. The FSF's insistence on "user freedom" moves beyond financial settlements, directly challenging the commercial viability of closed-source AI. If an AI company's business model relies on proprietary control over models trained on copyleft material, it may not be sustainable in the long run. This isn't merely a philosophical ideal; it's a direct call for a future where powerful AI models' foundational components are openly accessible, or face significant legal hurdles that could redefine how Anthropic LLMs are built, licensed, and deployed.