Anthropic Reaches $1.5B Settlement with Authors — The Largest AI Copyright Payout in History
Anthropic has agreed to a $1.5 billion settlement in the Bartz v. Anthropic class action, making it the largest AI copyright payout in history. Authors who are sole copyright owners receive approximately $3,000 per qualifying work; those sharing rights with publishers receive a 50/50 split of roughly $1,500. The settlement fund is paid in four instalments: $300M in October 2025, $300M post-approval, and $450M each in September 2026 and 2027. The claims deadline is March 30, 2026, with a Final Approval Hearing before Judge William Alsup on April 23. The case centred on Anthropic's use of books from shadow libraries LibGen and PiLiMi to train Claude.
Analysis
The $1.5 billion Bartz v. Anthropic settlement is the most consequential single event in the AI copyright story since the lawsuits began — not because of the money, but because of what Judge William Alsup's June 2025 summary judgment ruling established before the settlement was reached.
The ruling's core distinction is the one that will define AI copyright litigation for the next decade: training AI on legally acquired books may constitute fair use, but training on pirated materials obtained from shadow libraries does not. This is not a ruling about whether AI training is transformative, or whether it competes with the market for original works, or any of the other factors that have made fair use analysis in AI cases so contested. It is a ruling about the source of the training data. Anthropic's liability did not arise from what it did with the books; it arose from where it got them.
This distinction has immediate implications for every AI company that has used datasets assembled from LibGen, Z-Library, Books3, or similar sources. The legal risk is not theoretical — it is established. The question is no longer whether training on pirated data is infringing, but how much it costs. At $3,000 per qualifying work, the Anthropic settlement provides the first concrete answer to that question, and it is a number that every AI company's legal team is now modelling against its own training data exposure.
The Authors Guild's framing of the settlement as allowing authors to be "paid handsomely" is understandable from an advocacy perspective, but it deserves scrutiny. Three thousand dollars per work is a one-time payment for a use that will generate revenue for Anthropic indefinitely. It is not a licensing arrangement; it is a settlement of past infringement. Authors who accept the settlement waive their claims against Anthropic for the specific uses covered, but they retain their rights for future uses — and the settlement's transparency requirements, which mandate stricter data acquisition standards going forward, are arguably more valuable than the payment itself.
The March 30 claims deadline is urgent. Authors with qualifying works published before the class period cutoff should verify their eligibility immediately through the settlement administrator.