A Judge Signs Off on Anthropic's $1.5 Billion Copyright Settlement — the Largest Ever

# A Judge Signs Off on Anthropic's $1.5 Billion Copyright Settlement — the Largest Ever

A federal judge in San Francisco has given final approval to Anthropic's roughly $1.5 billion settlement with a class of authors, closing out the most expensive copyright case in U.S. history and setting a price on the pirated books that helped train one of the world's leading AI models.

U.S. District Judge Araceli Martínez-Olguín signed the final approval order on July 20, 2026, in Bartz v. Anthropic, finding the deal "fair and adequate" and overruling objections from a small group of dissenting authors. The ruling clears Anthropic to begin paying out roughly $3,000 for each of the approximately 500,000 works covered by the settlement — money owed to the writers and publishers who hold rights to books the company downloaded from pirate sites to build its Claude chatbot.

What the deal actually covers

The case was brought in 2024 by thriller novelist Andrea Bartz and nonfiction authors Charles Graeber and Kirk Wallace Johnson, who accused Anthropic of ingesting copies of their books without permission. Central to the dispute was where those books came from. Anthropic acquired large volumes of text from notorious "shadow libraries" — pirate repositories such as Library Genesis — rather than buying or licensing the titles.

Under the settlement, Anthropic will pay approximately $1.5 billion into a fund, destroy the original pirated files it torrented and stored, and compensate rightsholders on a per-work basis. Court filings put the payout at about $3,000 per book against a class of roughly 482,000 to 500,000 titles. The claims rate was unusually high: filings cited by author advocacy groups reported that more than 92 percent of eligible works had been claimed by the time of final approval.

In her order, Martínez-Olguín trimmed the plaintiffs' attorneys' fee request, awarding roughly $101.6 million rather than the $187.5 million the lawyers had sought — still an enormous fee, but a signal that the court scrutinized the terms rather than rubber-stamping them.

It is worth being precise about who did what. An earlier, closely watched fair-use ruling in the case came from Judge William Alsup, who oversaw the litigation before the settlement phase. Martínez-Olguín handled the settlement approval itself.

What was — and wasn't — decided about fair use

The headline number obscures a subtler legal reality. In 2025, Judge Alsup issued a split ruling on summary judgment: using books to train a large language model could qualify as fair use when the underlying copies were legally acquired, but Anthropic's decision to download and permanently store millions of pirated books was a separate act of infringement. It was that piracy — not the training itself — that exposed the company to potentially catastrophic statutory damages and drove it to settle.

Because Anthropic chose to pay rather than litigate to an appeals court, Alsup's fair-use analysis will not become binding precedent. As the Authors Guild noted in its guidance to members, the settlement resolves the money but leaves the central legal question unsettled for the next case.

Authors Guild CEO Mary Rasenberger framed the outcome as a market correction rather than a courtroom verdict. "We expect that the settlement will lead to more licensing that gives authors both compensation and control over the use of their work by AI companies, as should be the case in a functioning free market society," she said.

Why it matters

The number alone reshapes the risk math for the AI industry. A $1.5 billion payout — the largest known copyright settlement in U.S. history — establishes a rough market rate for training on unlicensed books and puts every other company facing similar suits on notice. Comparable litigation against OpenAI, Meta, Microsoft and others has been grinding through the courts, and plaintiffs now have a concrete benchmark to point to.

Just as important is what the deal preserves. Alsup's distinction between lawful training and unlawful acquisition offers a roadmap: the danger for AI labs may lie less in the act of training than in how they got the data. That could accelerate a shift already underway, in which companies strike licensing deals with publishers rather than scrape pirated corpora — changing the economics of building frontier models. Data that was effectively free becomes a line item, and the advantage tilts toward players who can afford to pay for it.

But the ambiguity cuts both ways. Because no appellate court has ruled, the fair-use question remains genuinely open. A future judge could weigh the four fair-use factors differently, and a different set of facts — say, a model that reproduces text verbatim — could produce a very different result. The settlement buys Anthropic peace and buys authors a payday, but it does not tell the industry where the legal line ultimately sits.

What to watch

The immediate story is administrative: how quickly the claims fund distributes money, whether any dissenting authors appeal the approval, and how publishers and estates divide payments with individual writers. Beyond that, the pressure moves to the other AI copyright suits, where plaintiffs will invoke the Anthropic figure and defendants will stress that a settlement sets no precedent. Watch, too, for a wave of licensing announcements — the clearest sign that the industry has decided paying up front is cheaper than being sued later.

---

Sources: [U.S. News](https://money.usnews.com/investing/news/articles/2026-07-20/us-judge-approves-anthropics-1-5-billion-settlement-of-copyright-lawsuit), [TechCrunch](https://techcrunch.com/2026/07/20/anthropics-landmark-1-5b-copyright-settlement-is-approved/), [Publishers Weekly](https://www.publishersweekly.com/pw/by-topic/digital/copyright/article/100888-judge-gives-final-approval-in-1-5-billion-settlement-in-anthropic-copyright-case.html), [Norton Rose Fulbright](https://www.nortonrosefulbright.com/en/knowledge/publications/ce8eaa5f/ai-in-litigation-series-an-update-on-ai-copyright-cases-in-2026).