AI Training on Books Faces Complex Copyright Battles and Uncertain Legal Ground

Copyright law, unchanged since 1976, forces judges to interpret old rules for modern AI training, creating a tangled legal environment. In a landmark case, Judge William Alsup imposed a $1.5 billion penalty on Anthropic for obtaining books from illegal shadow libraries, yet concluded that the act of training its language model was permissible, comparing it to a writer studying literature. Attorneys like Cathy Gellis argue that this outcome benefits AI companies, given the fine is modest relative to projected revenues.

The core question revolves around fair use and whether the AI's consumption of works is transformative enough to avoid infringement. Recent rulings, such as Judge Stephanos Bibas rejecting Ross Intelligence's use of Thomson one outlet content for a competing product, show courts drawing a line when the training directly challenges the source market. Meanwhile, the broader industry remains entangled in multiple lawsuits, leaving the ultimate legal standards for AI training uncertain.