> > AI training might be copyright infringement. But there’s no cases or laws to establish that.

> In September 2025, Anthropic agreed to pay at least $1.5 billion to settle a class-action lawsuit over using roughly 500,000 copyrighted books from "shadow libraries" to train their Claude LLMs.

Yes, but not because they were training LLMs with it. The judge in the case found specifically that training the LLMs on the copyrighted material was not copyright infringement; the only copyright infringement Anthropic had committed was acquiring the material itself. In other words, if they had legally bought all of the books they used, they would have been able to train their LLMs on them with no recourse from rights holders.