Yes, AI Can Train on Books: Anthropic Ruling

By Rebecca Grant: Yes, AI models can train on books.  It’s not a copyright infringement.

So says the opinion from Judge William Alsup, who ruled it was legal under the Fair Use doctrine for AI firm Anthropic to include published books in the data training sets for AI models.  In this case, Anthropic used published works to train its AI model named Claude. 

As it turns out, teaching large language models to generate human-like answers, dialogue, and essays takes a whole lot of practice with written work.  The June 24 ruling is the first affirmation that AI companies can have their models “read” or train on purchased books without violating copyright.

“Authors’ complaint is no different than it would be if they complained that training schoolchildren to write well would result in an explosion of competing works,” wrote Judge Alsup of the U.S. District Court for the Northern District of California.

The Fair Use Act of 1976 clarified occasions where it is okay to use a published work without violating copyright.  Some of the well-known instances include brief excerpts such as book reviews, and creating a parody. This ruling deemed the training of AI models to be a case of fair use.

Some warned the ruling was “more complicated than it looks” due to separate issues with pirated copies and the storage of published work in central libraries. “It’s a first-of-its-kind ruling in favor of the AI industry, but it’s importantly limited specifically to physical books Anthropic purchased and digitized,” noted Emma Roth at The Verge.  Many other AI and Fair Use cases are yet to be decided.

Still, the precedent set for allowing AI innovators to train models on published works has huge implications for U.S. national security and the technology race with China.  Over in China, the Cyberspace Administration of China uses “Xi Jinping thought” to train AI on how to control content on their internet. China has a chilling plan to lead the world in AI. The country’s AI industry and related sectors could grow into a market valued at $1.4 trillion by 2030, according to Morgan Stanley. Gaining global market share is step one for China’s plan.  American AI models can defeat this plan by continuing to develop the best products.

American AI leaders still have the skills to stay ahead of China’s advances, like DeepSeek, and other chatbots and generative AI models.  Advancing the leading edge is essential to retain market share – imagine a world where cheap Chinese “ChatXiPT” AI models squeezed out American products.  Judge Alsup’s common-sense ruling on AI training will help keep up the edge of excellence in American innovation.