Anthropic Settles Pirated Books Case for $1.5B

Anthropic has agreed to a proposed $1.5 billion settlement following allegations that it used pirated books to help train its AI model, Claude. If the settlement is approved, each eligible book title could receive compensation of around $3,000.

Key points:

The company has agreed to the proposed $1.5 billion settlement, with payments to be made if the court signs off on the plan.
The case highlights the growing legal risks of using unlicensed material in AI training, a move that may encourage more formal licensing agreements in the future.
For digital marketers, it’s a reminder that the legal basis of AI tools is an important factor to consider.

The settlement comes in the case of Bartz v. Anthropic, where the company was accused of downloading pirated books to train its AI systems. According to lawyers for the plaintiffs, if approved, this would represent the largest copyright settlement ever achieved in the United States. A preliminary approval hearing is scheduled to take place today.

Earlier in June, Judge William Alsup ruled that while training on books obtained legally can fall under “fair use,” the act of copying and storing millions of pirated books amounts to copyright infringement. That ruling paved the way for settlement negotiations, ultimately leading to this proposed agreement.

Settlement Details

Under the terms of the agreement, each qualifying book would receive around $3,000, with the class estimated to cover roughly 500,000 titles. The authors bringing the case allege that Anthropic obtained at least 7 million pirated copies from sites such as Library Genesis and Pirate Library Mirror.

Justin Nelson, one of the lawyers representing the authors, remarked that this appears to be “the largest copyright recovery ever achieved.”

How Payouts Would Work

The Authors Guild outlined that the settlement fund would be distributed in four stages, following court approval. This includes $300 million shortly after the preliminary decision, another $300 million once the final approval is granted, then $450 million after 12 months, and a further $450 million at 24 months. Interest will accrue while the money is held in escrow.

A finalised “Works List” is expected by 10 October, which will form the basis of a searchable database for claimants.

The Guild also pointed out that the deal requires the removal of pirated copies and applies only to past actions, not future conduct.

Why This Matters

For those who depend on AI tools in their content processes, the origin of training data is becoming increasingly important. It is likely that we will see more licensing agreements and greater transparency from providers about where their data comes from.

For publishers and authors, the payout per title establishes a benchmark that could provide stronger leverage when entering into future licensing negotiations.