Landmark Ruling: Court Permits Anthropic to Train AI on Copyrighted Books

Federal Judge William Alsup ruled that Anthropic was entitled to train its models on published books without the authors’ permission.

This marks the first instance where a court has upheld an AI company’s claim that the doctrine of fair use can exempt it from liability when using copyrighted materials to train large language models.

Previously, authors, artists, and publishers filed numerous lawsuits against AI developers. Although Judge Alsup’s decision does not ensure that other courts will adopt a similar stance, it could set a precedent favoring tech companies.

Decisions hinge on the interpretation of the fair use doctrine—a principle that permits the use of copyrighted materials without the rights holder’s permission in certain cases.

In the case against Anthropic, a group of plaintiffs accused the company of creating a “central library of all the books in the world” to store them “forever.” The works were allegedly downloaded illegally from pirate sites.

The judge stated that training neural networks with the materials was fair use, while the nature of the “central library” would be further examined.

“We will conduct a trial regarding the pirated copies used to create Anthropic’s central library and the associated damages. The fact that the company later purchased a copy of a book it had previously stolen from the internet does not absolve it from liability for theft, but it may affect the amount of statutory damages,” Judge Alsup wrote in his decision.

In June, the AI startup for image generation Midjourney became the latest company accused of violating intellectual property rights.