Anthropic trained Claude on illegally obtained Stanford professors' books

AI models may need to be retrained if they were trained on illegally obtained books, potentially impacting their capabilities and the costs for AI companies.

A federal judge ruled in May 2026 that AI company Anthropic illegally used copyrighted books from over 10 Stanford professors to train its Claude models. The lawsuit, Bartz v. Anthropic, resulted in a $1.5 billion settlement for authors and publishers, though some authors, including Professor Tobias Wolff, found the settlement insufficient and opted out. Wolff and other professors filed a separate lawsuit, Cruz v. Anthropic, seeking higher damages and a jury trial, alleging direct infringement through "shadow libraries." Judge William Alsup compared AI training to human learning, deeming it fair use for creating new content but blocking the creation of permanent libraries of pirated works. This ruling impacts the ongoing legal battles between authors and AI companies over data scraping and copyright infringement, with over 70 lawsuits filed against AI firms.

The ruling establishes a precedent that while AI training might be considered fair use for generative purposes, the systematic scraping and permanent storage of copyrighted works for a central library is unlawful. This distinction is critical for how AI companies source their training data moving forward. The ongoing legal challenges and the pursuit of statutory damages by authors like Wolff highlight the significant financial and legal risks AI developers face in data acquisition.

The Authors Guild's push for national legislation requiring licensing for AI training data, alongside ongoing lawsuits, suggests a tightening regulatory environment for AI development. This could lead to increased operational costs for AI companies if they are forced to license content, potentially slowing down development or shifting focus to different data sources. The demand for author attribution also points to a broader discussion about intellectual property rights in the age of AI.