The Anthropic AI copyright settlement is a $1.5 billion class-action deal that resolves claims Anthropic trained its Claude chatbot on hundreds of thousands of pirated books. Rightsholders of roughly 500,000 qualifying titles are set to receive about $3,000 per book, and Anthropic must destroy the pirated files it downloaded from Library Genesis (LibGen) and the Pirate Library Mirror (PiLiMi). A federal judge in California granted preliminary approval in September 2025. Final approval was still pending as of mid-2026 after the case was reassigned to a new judge.1Susman Godfrey. Susman Godfrey Secures $1.5 Billion Settlement in Landmark AI Piracy Case
What the Settlement Pays and Requires
The fund totals $1.5 billion plus interest and is non-reversionary, meaning unclaimed money stays with the class rather than going back to Anthropic. Each eligible title yields approximately $3,000 before legal fees and administrative costs. If the final list of qualifying works exceeds the anticipated 500,000 titles, Anthropic must add another $3,000 for each additional book.2Lieff Cabraser. Authors Secure $1.5 Billion Settlement in Landmark AI Piracy Case
Anthropic is paying in four installments: $300 million by October 2, 2025; $300 million within a week of final approval; $450 million by September 25, 2026; and $450 million by September 25, 2027.3Authors Guild. What Authors Need to Know About the Anthropic Settlement
How the per-book money splits depends on who holds the rights. Self-published authors, and authors whose rights have reverted, take 100% of the award for their title. For trade and university press books, the default split is 50/50 between author and publisher, though claimants can specify an alternative split based on their contracts.3Authors Guild. What Authors Need to Know About the Anthropic Settlement
On top of the money, Anthropic must destroy the original pirated files from LibGen and PiLiMi along with any copies derived from them. The company certified in the agreement that it did not use materials from those pirate libraries in any commercial models.2Lieff Cabraser. Authors Secure $1.5 Billion Settlement in Landmark AI Piracy Case Anthropic denies wrongdoing and maintains its use of the datasets was fair use.
Which Books Qualify
The certified class covers copyright owners of works found in the LibGen and PiLiMi datasets Anthropic downloaded. To qualify, a book needs an ISBN or ASIN and a U.S. Copyright Office registration that was timely, meaning made within five years of publication and either before Anthropic’s download date or within three months of publication.4Authors Alliance. Bartz v. Anthropic: Judge Alsup Certifies Class for Rightsholders of 7 Million Books
Books that appeared only in the “Books3” subset of the open-source dataset The Pile were excluded from the class because the metadata was insufficient to identify individual titles and authors. Judge William Alsup, who initially oversaw the case, certified a class encompassing roughly seven million titles found in LibGen and PiLiMi, though the settlement targets an anticipated 500,000 qualifying works after registration and other requirements are applied.
How Authors and Publishers Filed Claims
The settlement is administered by JND Legal Administration through AnthropicCopyrightSettlement.com, where rightsholders could search the official Works List and submit claims online, by email, or by mail.5AnthropicCopyrightSettlement.com. Bartz v. Anthropic Settlement FAQ The claim deadline was March 30, 2026.6AnthropicCopyrightSettlement.com. Bartz v. Anthropic Settlement Options and Due Dates
Both the author and the publisher of the same book could file separate claims. Penguin Random House, for example, told its authors that payment for a given title would be split 50/50 by default between the legal owner (typically the publisher) and the beneficial owner (typically the author), with each side filing its own claim.7Penguin Random House. Bartz v. Anthropic Copyright Settlement FAQ for Authors If co-owners disagreed on how to divide the award, the dispute could be referred to a court-appointed Special Master.
What the Settlement Does Not Cover
The release is narrower than the dollar figure might suggest. It covers only past claims for Anthropic’s torrenting, copying, retention, and use of works through August 25, 2025. It does not grant Anthropic any license to use the works for future AI training, and it does not release claims about the outputs Anthropic’s models generate.8ClassAction.org. Bartz v. Anthropic PBC, Settlement Notice Authors and publishers who believe Claude’s outputs infringe their work, or who object to any future training use, keep those claims.
Why Anthropic Paid: The Fair Use Ruling
The case turned on a summary judgment ruling Judge Alsup issued on June 23, 2025. He drew a sharp line between two kinds of copying. Training large language models on copyrighted works, he held, is “quintessentially transformative” and constitutes fair use as a matter of law. He reached the same conclusion about Anthropic buying print books and scanning them into a digital research library.9White & Case. Two California District Judges Rule Using Books to Train AI Is Fair Use
Downloading pirated copies was different. Judge Alsup ruled that copying books from LibGen and PiLiMi to build a general-purpose library was “inherently, irredeemably infringing.” He noted Anthropic had acquired at least five million copies from LibGen and two million from PiLiMi, and that it retained those files even when they were not actively being used for training.10Authors Alliance. Anthropic Wins on Fair Use for Training Its LLMs, Loses on Building a Central Library of Pirated Books With fair use knocked out for the pirated materials, the case was headed to trial on infringement and damages in December 2025. Statutory damages under the Copyright Act can reach $150,000 per willfully infringed work, a ceiling that gave the class enormous leverage. Anthropic settled instead.3Authors Guild. What Authors Need to Know About the Anthropic Settlement
Current Status: New Judge, Pending Final Approval
Judge Alsup granted preliminary approval on September 25, 2025 after initially rejecting the deal on September 8 without prejudice, saying he felt “misled” and demanding a definitive list of affected works and a revised claims protocol.11Authors Alliance. Bartz v. Anthropic: Updated Opt-Out and Objection Dates and a New Judge
In early 2026, Judge Alsup took inactive status and the case was randomly reassigned to Judge Araceli Martínez-Olguín, who extended the opt-out and objection deadline to January 29, 2026. She held the final approval hearing on May 14, 2026. Seven objectors presented concerns ranging from the size of the $3,000 payment relative to the $150,000 statutory ceiling to the treatment of pseudonymous, self-published, and unregistered-work authors.12Authors Alliance. Bartz v. Anthropic: Fairness Hearing Observations and Takeaways The judge did not rule from the bench and took the matter under submission, ordering Anthropic to file a supplemental brief by May 21, 2026 on whether late opt-outs should be honored.13Publishers Weekly. Little Drama at Anthropic’s Settlement Hearing As of the hearing, 92.77% of eligible class members had opted in.
A dispute over attorneys’ fees also remained unresolved. Co-lead class counsel Susman Godfrey and Lieff Cabraser requested approximately $225 million (15% of the fund) and proposed allocating another $75 million to three additional firms. In a December 23, 2025 memorandum, Judge Alsup called those additional firms “interlopers” and said he would not permit class counsel to divert any portion of its fee award to lawyers he had not approved. Anthropic separately opposed the add-on fees, arguing the amounts were 15 to 95 times what the firms said they had incurred.14Bloomberg Law. Anthropic Blasts Add-On Firms Bid for $75 Million of IP Deal
Authors Who Opted Out
Most eligible class members stayed in, but some notable authors chose to sue on their own. Journalist John Carreyrou, known for his Theranos reporting, and several co-plaintiffs opted out and filed a separate copyright infringement lawsuit on December 22, 2025 targeting Anthropic along with Google, OpenAI, Meta, xAI, and Perplexity AI.15Bloomberg Law. OpenAI, Anthropic, xAI Hit With Copyright Lawsuit From Writers The plaintiffs said they wanted to keep full control of their case rather than see their rights “diluted” in a class settlement, and pointed to the gap between $3,000 per work and the $150,000 statutory maximum for willful infringement. The case is before Judge Trina L. Thompson in the Northern District of California.16CourtListener. Carreyrou v. Anthropic PBC, No. 3:25-cv-10897
What It Means for AI and Copyright
More than 50 AI copyright lawsuits were pending across the country as of late 2025, and the $1.5 billion figure sent a clear signal about the cost of using pirated training data.17Copyright Alliance. Participating in the Bartz v. Anthropic Settlement The distinction Judge Alsup drew, that training on lawfully acquired material can be fair use but downloading from pirate sites is not, gave other courts and litigants a template. The $3,000-per-work payment, four times the $750 statutory minimum, became a benchmark in other negotiations. Music publishers including Universal Music Publishing, Concord, and ABKCO cited the Bartz ruling when moving to amend their own pending suit against Anthropic to add piracy-based claims tied to lyrics and sheet music alleged to be in the same LibGen and PiLiMi downloads.18Complete Music Update. Music Publishers Want to Amend Anthropic Lawsuit Because of Pirated Lyrics
Because the case settled instead of going to a jury, it produced no binding trial-level ruling on damages or the full scope of fair use in AI training. AI companies avoided a potentially devastating damages precedent; rightsholders lost the chance to establish one. What the settlement did establish, in dollars, is that licensing training data is cheaper than piracy.