Anthropic Agrees to Landmark $1.5 Billion Settlement, But Hurdles Remain

On Friday, AI company Anthropic announced that it had agreed to settle a copyright infringement lawsuit filed by authors who accused the platform of using pirated copies of their works to train its large language models (“LLMs”). The case has been closely watched given the rapid adoption of LLMs and their insatiable demand for training data.

While Anthropic secured a partial victory for LLM developers—Judge William Alsup ruled that using copyrighted data to train models constituted “fair use”—the win was not complete. Judge Alsup also found that Anthropic’s downloading of pirated copies of the authors’ works did not qualify as “fair use.”

As any experienced class action attorney knows, a settlement agreement is only the first step in a long, multi-stage process that takes months, if not years, to resolve. Here, it already appears the parties may face significant challenges in obtaining court approval. At a hearing on Monday, Judge Alsup expressed skepticism about the proposed settlement. His concerns included whether the claims process would adequately notify authors, whether large publishing groups might pressure authors into accepting the settlement, and the lack of clarity around the number of pirated works—currently estimated at 465,000 books. Judge Alsup wasted no time in setting deadlines for the parties to submit a list of the pirated works, a proposed claims form, and a follow-up hearing scheduled for the end of September.

The Judge’s concerns are only the beginning. The settlement will face additional scrutiny during the claims process, as Judge Alsup has already indicated he expects to see a high participation rate. For a settlement of this scale, a significant number of objectors and holdouts is almost inevitable. Depending on the final terms, any one of these obstacles could derail the agreement.