Judge Signs Off on Anthropic's $1.5 Billion Authors Settlement

A federal judge granted final approval on July 20, 2026 to Anthropic's $1.5 billion settlement with authors and publishers, closing the biggest copyright class action in history. Anthropic admitted no liability, and the court's earlier finding that training on lawfully acquired books is fair use still stands.

What happened

The case, Bartz v. Anthropic, began as a challenge to how the company assembled book data for training Claude. In 2025, the judge then overseeing it, William Alsup, drew a line that defined everything after: training a model on books the company had bought was fair use, but keeping a library of pirated copies was not covered. The pirated-library question was headed for trial, and statutory damages for willful infringement can reach very large numbers per work. Anthropic settled; preliminary approval came in September 2025.

Judge Araceli Martinez-Olguin, who took over after Alsup retired, granted final approval. Reporting puts the payout at roughly $3,000 to $3,100 per work across about 482,460 books. She noted that figure is about four times the minimum statutory damages amount for willful infringement, and rejected objections that the deal was too low. Anthropic said more than 91% of covered authors and publishers had claimed their share.

What the settlement does not cover

The release is narrower than the headline suggests. It covers the specific acquisition of the books at issue, not claims about what a model outputs. It also does not reach Anthropic's conduct on or after August 25, 2025. Some authors and publishers opted out and have separate suits pending. In other words, the deal buys peace on one set of past facts, not a licence to train on anything going forward.

That distinction explains why other AI developers watched closely. The fair-use finding for lawful training helps defendants, but it is a district court ruling and not binding elsewhere. The part of the case that cost Anthropic money was the sourcing of the data, not the act of training.

Why it matters

The settlement sets a price signal. About $3,000 a book is not a ceiling on future damages, but it gives plaintiffs in other training-data cases a concrete benchmark and gives defendants a reason to document where their data came from. It also pushes the industry toward licensing: obtaining content legitimately now looks cheaper than defending a trial with statutory damages hanging over it.

The court's decision to trim the fee request is a smaller signal worth noting. Judges are scrutinizing the economics of mega-settlements, and the lodestar approach implies courts will not simply hand lawyers a fixed share of a very large fund.

How the money moves

The arithmetic is straightforward. A fund of $1.5 billion spread over roughly 482,000 works comes to a little over $3,000 each before fees and costs, and the fee award of about $101 million, rather than the $187.5 million requested, leaves more in the pool for claimants. With more than 91% of covered authors and publishers filing claims, take-up is unusually high for a class settlement, where participation rates are often a small fraction of the class. Books were often co-owned by authors and publishers, so each work's payment is typically divided between them under the claims process.

The judge's reasoning on fees is a useful data point for future cases. By applying the lodestar method, which tallies hours and rates, rather than a percentage of the fund, she signalled that the sheer size of a settlement does not automatically justify a proportional fee. In megafund cases, that approach can shave tens of millions off a request.

For Anthropic, the deal removes a trial that carried large and unpredictable exposure. Statutory damages for willful infringement run per work, and with hundreds of thousands of titles in play, a loss could have been multiples of the settlement. The company's position remains that the underlying fair-use ruling on training is still the law in that court. That lets it argue it paid to resolve how it obtained data, not for the right to train, a framing other labs will lean on.

What to watch

• Opt-out cases against Anthropic, which could test claims the class release left behind.
• Whether the fair-use reasoning is adopted or rejected in appeals involving other AI companies.
• More licensing deals between publishers and AI labs, now that a benchmark exists.
• The still-pending newspaper cases against OpenAI, which turn on different facts.

Sources

1. Anthropic's $1.5 billion copyright settlement receives final court approval — BetaNews
2. Anthropic $1.5 Billion Copyright Pact Wins Judge's Approval — Bloomberg Law
3. Judge Approves Anthropic's $1.5 Billion Copyright Settlement With Authors — Digital Citizen
4. US judge approves Anthropic's $1.5 billion copyright settlement with authors — CADE Project