Anthropic Settles Copyright Lawsuit for 1.5 Billion Dollars

AI startup Anthropic has agreed to a 1.5 billion dollar settlement in a class action lawsuit over book copyright infringement. This is considered the largest settlement amount in the copyright field for class actions. While it may appear at first glance to be a major victory for authors, analyzing the structure of the settlement reveals that it is substantively favorable to AI laboratories.

At the core of the lawsuit is the question of what data was used to train AI. The settlement is positioned as compensation for Anthropic's download of approximately 482,460 books from pirated databases. In other words, the payment target is limited not to "the act of AI learning itself" but to "the act of unlawfully obtaining copyrighted works." This distinction significantly affects the implications for the entire industry.

Importantly, the presiding judge's judgment prior to the settlement must be noted. The judge expressed the view that using legitimately obtained books in AI training constitutes "transformative use" and falls within the scope of copyright law's fair use doctrine. Fair use is an exception that allows the use of copyrighted works without the copyright holder's permission, and may apply to uses that create new meaning or value. This judgment forms the foundation for interpreting this settlement.

To summarize, Anthropic was forced to pay a substantial amount not because of "teaching books to AI" but because of "illegally obtaining data from pirated websites." These are similar but distinct issues; the latter is an illegal act that can be questioned in any industry. This distinction is significant in showing the future direction of legal discussions surrounding AI and copyright.

Multiple lawsuits are currently ongoing in the AI industry over copyright issues related to training data. Within this context, this settlement can be positioned as a potential legal foundation for AI companies in that it "did not undermine the principle that data properly obtained can be used for learning." While paying a substantial settlement amount, Anthropic did not concede on the core issue of "the legality of learning conduct," which was most disputed.

For authors, the 1.5 billion dollar settlement is also not insignificant. However, since the original goal of having AI learning itself declared illegal was not achieved, some may view it as a limited victory from the perspective of copyright protection. The legal interpretation in this field may change further depending on what evidence and logic are used in future lawsuits.

The greatest issue highlighted by this settlement is the reality that rules governing AI training data have not yet been established. How to combine "the legality of data acquisition" (where it came from) with "the nature of its use" (what it is used for) is likely to become the focus of future litigation and legal reform. As generative AI becomes more integrated into society, the demand for answers to this question will only grow.