US judge rules Anthropic’s use of books for AI training is fair use

A federal judge in the United States has delivered a pivotal ruling regarding Anthropic's use of copyrighted literature for training its artificial intelligence system, declaring it as fair use. However, the judge also found the company liable for copyright infringement due to its storage of pirated digital copies of millions of books. This significant decision was made by District Judge William Alsup in San Francisco and comes amidst a series of legal challenges concerning how AI companies utilize copyrighted materials. The lawsuit, initiated by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, accuses Anthropic of employing unauthorized versions of their works to develop its Claude large language model. Judge Alsup ruled in favor of Anthropic regarding the central claim, emphasizing that the company’s utilization of the books during AI training was "exceedingly transformative," thus qualifying for protection under the fair use doctrine. He elaborated that Anthropic’s approach was not to replicate or overshadow these works but to innovate and create something entirely new. However, the judge did not overlook the company's actions concerning its so-called "central library," which housed over seven million pirated books. He determined that this conduct surpassed the boundaries of fair use and constituted copyright infringement. A jury trial has been scheduled for December to assess potential damages, which can soar up to $150,000 per work under U.S. copyright law. Anthropic, supported by tech giants Amazon and Alphabet, has not yet commented on this ruling. In previous legal documents, the company maintained that its AI training practices were legally sound and fostered innovation. Furthermore, Anthropic argued that the origins of the training data—whether legitimate or pirated—should not impact the fair use evaluation. Judge Alsup, however, rejected this perspective, casting doubt on the necessity of using pirated materials. He remarked, "This order doubts that any accused infringer could ever meet its burden of explaining why downloading source copies from pirate sites that it could have purchased or otherwise accessed lawfully was itself reasonably necessary to any subsequent fair use." This ruling marks a notable moment as it is the first instance where a court has specifically addressed the fair use defense in relation to generative AI, a legal domain still in its infancy. The case underscores the increasing friction between copyright holders and AI developers concerning the sourcing and application of creative works in machine learning. The upcoming trial will further shape the landscape of copyright law in the context of artificial intelligence, determining the financial repercussions for Anthropic's unauthorized storage of protected material.