In a book-lined office in central London, an editor at Bloomsbury Publishing reviews the latest royalty statement with quiet satisfaction. The numbers reflect more than routine sales. They trace back to a decision made thousands of miles away in a California courtroom, where a federal judge has now put the final seal on one of the largest copyright settlements involving artificial intelligence training data.
A US federal court granted final approval in July 2026 to a $1.5 billion settlement in the Bartz v Anthropic class action. The agreement resolves allegations that the AI company obtained copyrighted books from pirated sources, including sites such as LibGen, and used them to develop its Claude models. Bloomsbury has confirmed that 14,087 of its titles fall within the settlement class. Each qualifying title carries an allocation of roughly $3,000 before deductions, to be divided equally between author and publisher once fees are met.
Payments will arrive in several instalments, beginning in the second half of the financial year. For a house known for literary fiction, academic works and children’s books, the sum represents tangible recognition of value extracted without permission. The per-title figure, a judge observed, stands at approximately four times the statutory minimum, a detail that underscores the scale of the resolution while noting the litigation risks that made settlement prudent.
Market accountability without halting progress
The outcome arrives at a moment when rapid expansion of large language models has tested the boundaries of intellectual property. Rather than override established norms, Anthropic and the class of rights holders have reached terms that compensate creators and publishers. No admission of liability accompanied the payment, yet the size of the fund sends a clear signal: data scraped from unauthorised sources carries a price.
Bloomsbury’s participation highlights the reach of British publishing into the heart of these disputes. With titles spanning genres that shape cultural conversation on both sides of the Atlantic, the publisher’s catalogue forms part of the bedrock on which AI systems have been built. The settlement converts that contribution into direct financial return. Authors, who share the proceeds, gain practical benefit at a time when many worry about the devaluation of their work in the age of generative tools.
Viewed from the perspective of rights holders, the case demonstrates that legal mechanisms grounded in property principles can temper the ambitions of powerful technology firms. The court’s approval follows preliminary clearance in September 2025, a fairness hearing earlier this year, and the passage of opt-out and objection deadlines. Final approval on or around 21 July 2026 closes a chapter that began with authors alleging systematic infringement.
