Anthropic will pay at least 1.5 billion dollars for its use of pirated books to train Claude. A major step forward, but major legal questions about AI and copyright remain. The legal soap opera pitting authors and book publishers against the Anthropic company in the United States has just reached a major turning point. Just over two months after the decision adopted by Judge William Alsup, which recognized the applicability of fair use to the training of artificial intelligence models with legally acquired books while excluding pirated works from the scope of this exception, Anthropic has chosen to negotiate. The company, creator of the Claude language model, announced on September 5 that it had reached a settlement agreement with representatives of authors and publishers now grouped in the class action Bartz v. Anthropic. The settlement calls for the payment of at least $1.5 billion to repair the massive use of copies from pirate libraries such as LibGen or PiLiMi, as well as the destruction of all pirated files used to train Claude. Judge Alsup has yet to approve the settlement, with a preliminary…
