Anthropic $1.5 Billion Copyright Settlement Receives Final Court Approval - Featured image
Illustration for Anthropic $1.5 Billion Copyright Settlement Receiv

Anthropic $1.5 Billion Copyright Settlement Receives Final Court Approval

Historic $1.5 billion settlement receives court approval, compensating authors whose books were pirated to train AI.

On July 20, 2026, U.S. District Judge Araceli Martínez-Olguín granted final approval to the $1.5 billion class action settlement in Bartz v. Anthropic, marking the largest copyright settlement in American legal history. The decision ended nearly two years of litigation that began in August 2024 when nonfiction authors Charles Graeber (The Good Nurse), Kirk Wallace Johnson (The Feather Thief), and thriller author Andrea Bartz (We Never Here) sued the AI company for downloading approximately 7 million pirated books from shadow libraries to train its Claude AI models.

With final court approval now in place, the settlement moves into its claims distribution phase, making real compensation available to hundreds of thousands of affected authors. The settlement’s significance extends beyond its financial scale. The case established a crucial legal distinction: while using legally purchased books for AI training constitutes fair use, acquiring those same books through piracy does not. This ruling, confirmed in a June 2025 summary judgment by Judge William Alsup, transformed the legal landscape for how companies can source training data and directly led to Anthropic’s decision to settle rather than continue defending its data practices in court.

Official resources:

Table of Contents

Anthropic faced mounting legal exposure after Judge Alsup’s June 2025 ruling that downloading copyrighted books without authorization from Library Genesis and Pirate Library Mirror constituted copyright infringement despite the company’s fair use defense. The company had sourced approximately 7 million pirated copies from these shadow libraries to train its Claude models, a practice the judge determined fell outside fair use protections specifically because the source material was obtained illegally rather than purchased. With class certification already granted in August 2025 to cover roughly 500,000 affected book titles, Anthropic calculated that fighting the case to trial carried substantial financial and reputational risk. The settlement reflected a pragmatic calculation on both sides.

Anthropic avoided potential damages that could have exceeded the settlement amount, while authors gained immediate certainty of compensation rather than waiting years for trial outcomes. For reference, the $1.5 billion settlement dwarfs previous landmark cases—the Oracle v. Google Java API case, which received far more public attention, involved only $100 million in damages. This settlement’s scale underscores the growing scrutiny of AI companies’ data sourcing practices and signals that using pirated materials for training is now legally and economically indefensible.

The $1.5 billion payout represents an unprecedented amount in copyright litigation, reflecting both the scope of the infringement and the settlement‘s ambition to fairly compensate creators across an entire industry segment. By comparison, music licensing settlements and other major copyright cases have rarely exceeded a few hundred million dollars, making this settlement genuinely historic. Judge Martínez-Olguín’s approval process involved scrutinizing not just the overall amount but also how the money would be distributed, the fairness to class members, and the reasonableness of attorney fees.

Judge Martínez-Olguín awarded attorneys $101,561,111—a significant reduction from the requested $187.5 million. She applied a 3.75 multiplier to the lodestar (the actual hours worked multiplied by a reasonable hourly rate) but expressed concern that even this multiplier was steep for mega-fund cases. This decision acknowledged that while the attorneys performed substantial work, the multiplier needed to reflect the reality that large settlements command different market dynamics than smaller ones. The resulting fee—roughly 6.8% of the total settlement—remains substantial but demonstrates judicial skepticism toward legal fee inflation in record-breaking cases.

How Many Authors Are Covered and Who Qualifies?

The class encompasses approximately 500,000 book titles that were found in the LibGen and PiLiMi datasets that Anthropic downloaded. Of these, 447,576 works were actually claimed by authors or their estates, representing a 92.77% claim rate—an unusually high participation rate for class action settlements. This concentration suggests the settlement notice reached authors effectively and that eligible claimants recognized the value of participating. However, not every author whose work was pirated necessarily claimed compensation, whether due to lack of awareness, difficulty navigating the claims process, or other barriers.

Eligibility criteria are strict and require authors to meet multiple conditions. A work must have an International Standard Book Number (ISBN) or Amazon Standard Identification Number (ASIN), be registered with the U.S. Copyright Office within five years of publication (and before Anthropic’s download or within three months of publication), and appear on the certified Works List maintained by the settlement administrator. These requirements exclude certain legitimate books, such as self-published works without ISBNs, works published outside the United States, and books registered late. Authors must also affirmatively claim their works; the settlement does not automatically identify and pay claimants.

What Will Authors Actually Receive in Compensation?

Individual authors receive approximately $3,000 to $3,100 per claimed work after deduction of attorney fees, claims administration costs, and settlement expenses. This per-work payout was calculated by dividing the net settlement fund by the number of claimed works. For authors with multiple books in the pirated dataset—not uncommon for prolific nonfiction and fiction writers—the compensation can total tens of thousands of dollars. An author who had five books in the LibGen collection and successfully claimed all five would receive roughly $15,000 to $15,500, for example.

The per-work amount reflects a compromise between ensuring meaningful compensation to individual authors and avoiding an impractically small payout that would cost more to distribute than the amount paid. The settlement prioritized broad coverage of the class over maximizing individual payouts. Authors with fewer works or those whose books had lower commercial value at the time of infringement received the same per-work amount as bestselling authors, which some have criticized as undervaluing high-traffic books that likely contributed more to Claude’s training effectiveness. Claims distribution is expected to begin in late 2026, with payments processing over several months.

What Are the Eligibility Challenges and Timing Requirements?

Authors claiming compensation must satisfy particularly strict timing requirements around copyright registration. For most works, registration must have occurred within five years of publication, but also crucially either before Anthropic’s download from the shadow library or within three months of the publication date. This dual requirement creates a trap for authors with older works—a book published in 2010 and registered in 2014 might fall outside the acceptable window if Anthropic downloaded it before 2014 from LibGen. Authors whose books were published internationally but never registered with the U.S.

Copyright Office are ineligible, despite potentially having had their works pirated and used. The most significant limitation affects self-published authors and those working with small independent presses. Works must have either an ISBN or ASIN, requirements that exclude many self-published titles, particularly those released before ISBNs became standard for e-books. Additionally, the requirement to affirmatively claim works means that if an author fails to discover they’re eligible, fails to submit a claim, or submits incomplete documentation, they receive nothing. The settlement’s claims administrator has provided assistance, but navigating these technical requirements remains the author’s responsibility, introducing a substantial gap between theoretical eligibility and actual compensation.

The Destruction Requirement and Data Handling

The settlement imposes a critical obligation on Anthropic: the company must destroy all original files of works it obtained from Library Genesis and Pirate Library Mirror, along with any copies or derivatives created from those pirated materials, subject to legal preservation obligations. This requirement addresses a core concern among authors and copyright advocates—that Anthropic would retain pirated data indefinitely, continuing to benefit from the infringing material even after paying damages. The destruction provision attempts to terminate Anthropic’s ongoing use of the pirated training data.

However, the destruction requirement includes an important caveat: Anthropic may retain materials as required by law, regulation, or court order, and retains the right to preserve materials for legal proceedings or regulatory compliance. This exception is standard in settlement agreements but creates a potential loophole. If Anthropic argues that regulatory requirements or potential future litigation justify retention, the destruction requirement might not be enforced as strictly as intended. Additionally, destruction of “derivative” works is inherently difficult to verify—if Anthropic’s models incorporated knowledge from pirated books into their neural network weights, no external auditor can definitively confirm that all derived information has been eliminated.

The Fair Use Ruling That Changed Everything

Judge William Alsup’s June 2025 summary judgment ruling established a legally binding distinction that directly shaped this settlement. The court ruled that Anthropic could not claim fair use protection for downloading books from piracy sites, even though using legally purchased copies for AI training might qualify as fair use. This distinction—between using books you’ve lawfully obtained and sourcing books through illegal channels—created an insurmountable legal problem for Anthropic’s defense. The ruling acknowledged that in a different factual scenario, where a company licensed books legally or purchased them through legitimate channels, fair use arguments might prevail.

But Anthropic’s choice to source from shadow libraries transformed the legal analysis entirely. The August 2025 class certification by Judge Alsup defined the class as “all beneficial or legal copyright owners of the exclusive right to reproduce copies of any book” appearing in the LibGen or PiLiMi datasets, giving concrete shape to the group that would be eligible to recover. This certification was itself noteworthy—establishing a class action in copyright infringement cases presents technical challenges, and Judge Alsup’s decision reflected a determination that common issues predominated over individual ones. The court essentially found that Anthropic’s conduct toward all pirated books was sufficiently uniform that resolving the case on a classwide basis served judicial economy and fairness better than requiring hundreds of thousands of individual lawsuits.

  • —

You Might Also Like

Browse more: Class Action | showcase | Uncategorized