Federal Judge William Alsup has ruled that Anthropic’s training of its AI models on published books without obtaining permission from the authors is legal. This ruling is significant as it marks the first instance where a court has recognized that the fair use doctrine may protect AI companies from liability while using copyrighted material for training large language models (LLMs).
This decision poses challenges for authors, artists, and publishers who have filed numerous lawsuits against companies such as OpenAI, Meta, and Google. The implications of this ruling could be far-reaching, as it provides a potential legal precedent that may encourage other courts to favor technology companies over creative professionals.
These lawsuits typically hinge on the interpretation of the fair use doctrine, which has been criticized for its ambiguous nature and has remained unchanged since 1976, long before the emergence of the internet and generative AI. Fair use assessments consider the purpose of the use, potential commercial benefit, and how transformative the derivative work is compared to the original.
Other companies, like Meta, have also argued similar fair use claims to justify their use of copyrighted works for AI training, but prior to this ruling, the outcome was uncertain. In the specific case of Bartz v. Anthropic, plaintiff authors raised concerns regarding how Anthropic obtained and stored its materials.
The lawsuit alleges that Anthropic aimed to create a “central library” containing “all the books in the world,” but that many of these works were downloaded illegally from pirate websites. While Judge Alsup acknowledged that Anthropic’s use of the materials could be considered fair use, he emphasized that a trial would still be necessary to examine the nature of the library created and the associated damages.