Federal Judge Halts Anthropic $1.5B Authors Settlement

A federal judge has put on hold Anthropic’s recently announced $1.5 billion settlement with US authors who say the AI company trained models on pirated copies of their books. At a hearing, Judge William Alsup rejected the deal for now, saying he’s concerned class-action lawyers might negotiate terms “behind closed doors” and then force the outcome “down the throats of authors.” He also asked for more detail about the settlement’s claims process and warned of “hangers on with all this money on the table.”


Under the pact, authors and publishers would get roughly $3,000 per covered work; plaintiffs’ counsel estimated about 465,000 books could be eligible. Alsup previously ruled that using purchased books for training can be fair use but left open liability for models trained on illegally downloaded works — a distinction central to the case. He will revisit the settlement at a hearing on September 25, saying, “We’ll see if I can hold my nose and approve it.”


Industry figures are split: Maria Pallante of the Association of American Publishers criticized Alsup’s stance as misunderstanding publishing, while plaintiffs’ lawyer Justin Nelson said the attorneys are committed to compensating every proper claim.