Settlement reached in Bartz v. Anthropic over AI training data

A major class-action settlement in a high-profile AI copyright case was reached recently, ending a dispute over whether an AI company's training of its language model used authors’ books without permission. Anthropic agreed to a settlement with a group of authors who alleged that millions of works were pirated or downloaded to train Claude. The terms remain confidential, and final agreement is expected to be confirmed in the coming weeks, with a formal approval process moving forward this fall. The case, Bartz v. Anthropic, has been one of the most consequential tests of how copyright law applies to artificial intelligence and the training of large language models.

The legal battle centered on two intertwined questions: whether an AI developer’s use of copyrighted material to teach its models qualifies as fair use, and whether the means by which some works were acquired constitutes piracy that could trigger damages. In a ruling issued in June, a federal judge largely sided with Anthropic on the fair-use question, concluding that the training process transformed the works sufficiently to be legal. However, the judge also found that some of the books were obtained through shadow...