Coup de théâtre! A Californian judge has just ruled that Anthropic’s LLM Claude’s training with copyrighted works benefits from the « fair use » exception, provided the works were acquired legally. On the other hand, the AI developer’s piracy of books is not covered by the exception, and must be compensated for by the injured authors, a compensation that could reach several billion dollars. The application, or otherwise, of the fair use exception, or « fair use » to the training of generative AI with copyrighted works is at the heart of the debate, with nearly forty lawsuits in the United States, and more and more lawsuits in Europe too. Last February, a ruling in Delaware had denied the application of this exception to copyright to the use of protected content belonging to Thomson Reuters by the legal research AI developed by the Ross company. In a « summary » judgment issued on June 23, San Francisco judge William Alsup went in the opposite direction, concluding that Anthropic should benefit from the « fair use » exception for training its LLM Claude with copyrighted…
