A federal judge has ruled that Anthropic’s use of books to train its Claude AI model is partially legal under the fair use doctrine.

US District Judge William Alsup found the AI training “exceedingly transformative,” comparing it to how a reader learns to write through study.

However, he ruled that Anthropic infringed copyrights by copying and storing over 7 million pirated books in a “central library.”

A trial in December will determine the damages.

The lawsuit was filed as a proposed class action by authors Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson.

🔗 Source: Reuters

🧠 Food for thought

1️⃣ First major application of fair use to AI training marks pivotal moment in copyright evolution

Judge Alsup’s ruling represents the first significant judicial interpretation of fair use in generative AI, continuing a centuries-long pattern of copyright law adapting to technological change.

The decision specifically describes AI training as “exceedingly transformative,” drawing parallels to how human writers learn from reading others’ works 1, establishing a new judicial framework for evaluating AI’s use of copyrighted material.

This ruling follows the historical pattern seen when copyright law adapted to previous technologies, such as player pianos (addressed in White-Smith Music v. Apollo in 1908) and software in the 1960s-70s 2, where courts and legislators had to determine how existing principles apply to novel technologies.

The judge’s determination that AI models train on works “not to race ahead and replicate or supplant them, but to turn a hard corner and create something different” 1 establishes a new legal interpretation of “transformative use” that could influence copyright jurisprudence for decades.

Fair use itself has evolved significantly since being codified in the 1976 Copyright Act 3, with this ruling potentially becoming as influential as landmark cases like Folsom v. Marsh (1841) that established the foundations of fair use doctrine.

2️⃣ Critical distinction between training and storage creates complex compliance requirements

Judge Alsup drew a significant boundary between permissible AI training and impermissible storage of copyrighted works, creating a nuanced framework for how AI companies must handle content.

While ruling that training was fair use, the judge explicitly rejected Anthropic’s defense for storing 7 million pirated books in a “central library,” stating that downloading from pirate sites when legal access was available couldn’t be justified 4.

Recent Anthropic developments