Anthropic agreed to a settlement with Bloomsbury Publishing after using copyrighted books to train its Claude AI model [1, 2].
This agreement marks a significant legal pivot in the ongoing conflict between generative AI developers and the publishing industry. It establishes a financial precedent for how AI companies must compensate authors and publishers when using intellectual property for machine learning training.
The settlement follows a class-action copyright lawsuit alleging that Anthropic used pirated copies of books to develop its technology [1, 2]. Bloomsbury, the publisher of the Harry Potter series, is among the entities receiving payment for the unauthorized use of its catalog [1].
Reports on the total settlement amount vary across sources. The Guardian reported the total at $1.5 billion [1], while another report cited a figure of $2 billion [2]. UK-based reports show further discrepancies, with figures ranging from £14 million [3] to £1.1 billion [4].
The settlement covers 14,087 titles from the Bloomsbury catalog [1]. According to reporting from The Next Web, the approximate payout amounts to $3,000 per work [5].
Anthropic is a U.S.-based AI startup, while Bloomsbury is based in the UK [1, 2]. The resolution of this case comes as more publishers seek to protect their intellectual property from being ingested by large language models without consent or compensation.
“Anthropic agreed to a settlement with Bloomsbury Publishing after using copyrighted books to train its Claude AI model.”
This settlement signals a shift from the 'fair use' defense to a licensing model for AI training data. By paying billions of dollars to a major publisher, Anthropic acknowledges that the scale of data ingestion in modern AI requires a formal commercial framework, likely forcing other AI labs to negotiate similar deals to avoid protracted litigation.
