A US federal judge has determined that Anthropic’s utilization of books to train artificial intelligence (AI) tools without the consent of authors falls under “fair use” laws. This ruling comes amid the ongoing debate concerning the influence of AI on the creative industry and the industry’s efforts to promote relaxed regulatory measures.
US District Judge William Alsup stated that Anthropic’s AI models, rather than racing to replicate the works, aim to create something unique. Despite a group of authors claiming that Anthropic’s actions were unlawful, the judge concluded that the AI system did not breach US copyright laws, which intend to foster creativity and scientific progress.
The AI’s output was deemed by Alsup as “exceedingly transformative,” qualify it for “fair use” protection. However, Alsup did rule against Anthropic’s storage and copying of approximately seven million pirated books in a “central library,” as this was deemed to infringe upon author copyrights and did not meet the requirements of fair use.
This fair use doctrine is frequently employed by tech companies as they create generative AI, involving the compilation of existing content to train AI models. The ongoing argument questions whether AI will enhance artistic creativity or merely promote the mass production of affordable replicas, potentially rendering artists unemployed, to the advantage of major corporations.
The lawsuit was filed by Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson, who accused Anthropic of “large-scale theft” and exploiting the “human expression and ingenuity” encapsulated in each work to reap profits. Although Tuesday’s decision was celebrated by AI developers, the judge mandated that Anthropic would still face a trial in December regarding the alleged misuse of pirated material.
Judge Alsup stated that Anthropic held “no entitlement to use pirated copies for its central library.”
Source: https://www.aljazeera.com/news/2025/6/24/us-judge-allows-company-to-train-ai-using-copyrighted-literary-materials?traffic_source=rss
