Artificial-intelligence start-up Anthropic has reached a settlement with a group of U.S. authors who alleged the company infringed their copyrights by training its Claude language models on millions of pirated books, according to a filing on Tuesday with the Ninth Circuit Court of Appeals. The parties signed a binding term sheet on Aug. 25 and asked the court to pause proceedings while they finalize the agreement. U.S. District Judge William Alsup ordered the authors and Anthropic to seek preliminary approval of the deal by Sept. 5, and the December 2025 trial that had been set to determine damages has been taken off the calendar. Authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sued last year, claiming Anthropic illegally downloaded as many as seven million titles from shadow libraries. Statutory damages for willful infringement can reach $150,000 per work, exposing the Amazon- and Alphabet-backed company to potential liabilities in the billions. In June, Alsup ruled that training AI models on lawfully purchased books was fair use but allowed the case to proceed on allegations that Anthropic stored pirated works in a central database. He later certified what could have been a class of up to seven million authors, increasing the stakes for the San-Francisco-based firm. Settlement terms were not disclosed. Justin Nelson, counsel for the authors, called the accord “historic” and said details will be announced in the coming weeks. Anthropic declined to comment. Legal scholars say the deal could influence similar copyright disputes pending against OpenAI, Microsoft and Meta.