A federal court in San Francisco has approved a record $1.5 billion settlement agreement between Anthropic and a group of authors in a copyright infringement case. The lawsuit was filed in 2024, accusing them of using pirated copies of literary works to teach the Claude language model. Earlier, Judge William Alsup ruled that the very use of texts for teaching falls under the concept of fair use, but the storage of seven million unlicensed books in the company's central library was considered a violation. Initially, the judge rejected the proposed deal, expressing concerns that its terms put unnecessary pressure on the authors and left a number of important legal issues unresolved.

In response to the court's comments, Anthropic promptly launched a specialized web resource containing a complete list of affected works. This allowed the copyright holders to accurately determine whether their work falls under the terms of the agreement, which in total covers more than 480,000 titles. District Judge Araceli Martinez-Olgin, who took over the case after the previous judge resigned, finally approved the terms of the settlement. According to the document, the authors will receive compensation in the amount of approximately $ 3,000 for each affected work, and the company Anthropic has committed to completely destroy all unlicensed copies of books stored in its systems.

Despite the magnitude of the agreements reached, the reaction in the creative community turned out to be mixed. A number of plaintiffs considered the proposed amount of compensation not high enough to cover the actual damage caused by the use of their intellectual property. In this regard, some of the authors officially refused to participate in the collective agreement, reserving the right to file individual lawsuits against the developers of artificial intelligence. This precedent highlights the ongoing legal difficulties in regulating the training of neural networks on copyrighted materials.