Bartz v. Anthropic: The Fair Use Ruling and the Settlement

Who is suing and why

Authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sued Anthropic in August 2024 as a putative class action, alleging that the company pirated copies of their books for its library and reproduced them to train its large language models. Judge Alsup’s June 23, 2025 order (Dkt. 231) recounts the record: Anthropic downloaded millions of books from pirate sites, including Library Genesis and Pirate Library Mirror, and separately bought print books, stripped their bindings and scanned them into digital files. Both sets went into a central library from which training sets were drawn.

The case is one of several brought by authors against AI developers in the Northern District of California. Its importance comes from timing: it produced the first substantive fair use ruling on generative AI training, two days before Judge Vince Chhabria ruled in the parallel case against Meta, and it was the first of these cases to settle as a class. Later filings name corporate class representatives as well, including Andrea Bartz, Inc. and MJ + KJ, Inc.

Claims

Training on purchased books

The authors argued that every copy Anthropic made to train Claude was an unlicensed reproduction, and that the outputs of a model trained on their books would compete with them. Anthropic argued that training is a transformative use under 17 U.S.C. § 107, because the model learns statistical relationships in text rather than storing books to hand back to users. The order records that the authors conceded training produced no exact copies or infringing knockoffs of their works for the public, and said a case with such facts would be different.

Pirated library copies

The second claim targeted acquisition rather than training. Anthropic had downloaded pirated books and kept them, the order notes, as a permanent general-purpose library, including copies it later decided it would never use for training. The authors argued that this was infringement regardless of what the books were later used for. Anthropic argued that the downloads should be judged together with the training they served, so that a fair use for training would cover them too.

Key rulings

June 2025 anthropic fair use ruling

On summary judgment, Judge Alsup split the uses. Training was fair use: he described the use of the books to train Claude as “exceedingly transformative,” comparing the models to “any reader aspiring to be a writer.” He also rejected the argument that competing AI-written works were a cognizable harm, comparing it to complaining that teaching schoolchildren to write well would produce competing works. Scanning purchased print books was also fair use, for a narrower reason: Anthropic replaced the print copies it bought with more convenient digital ones, without adding copies or redistributing them. The pirated library was not. The order says that pirating copies otherwise available for purchase is “inherently, irredeemably infringing” and that every factor pointed against fair use for the downloaded library copies. It set a trial on those copies and on damages, including willfulness.

The Authors Guild has noted that the training ruling applied only to the three named plaintiffs, because the class later certified covered the piracy claim only (Authors Guild explainer). On July 17, 2025, Judge Alsup certified a class of rightsholders of books Anthropic downloaded from LibGen and PiLiMi that had an ISBN or ASIN and a timely copyright registration. Norton Rose Fulbright reported that Anthropic’s requests for an interlocutory appeal and a stay were denied in early August 2025, leaving trial set for December.

Bartz v anthropic settlement approval

The parties filed a $1.5 billion settlement on September 5, 2025, and Judge Alsup granted preliminary approval at a hearing on September 25, 2025, according to the Authors Guild’s update log. After Alsup retired, the case was reassigned to Judge Araceli Martínez-Olguín, who held the fairness hearing on May 14, 2026 and granted final approval on July 20, 2026. The final approval order (Dkt. 680) found the non-reversionary fund fair and adequate, quoted the earlier finding that the estimated award of about $3,000 per work is four times the statutory minimum for ordinary infringement, cut class counsel’s fee to $101,561,111, entered final judgment and dismissed the case with prejudice.

The release covers past conduct only, through August 25, 2025, and does not release claims over AI outputs. The settlement also requires Anthropic to destroy its original LibGen and PiLiMi files, subject to legal preservation duties. Payment details are on the settlement explainer and payout pages in the related links.

How Bartz differs from Kadrey v. Meta

Two days after the Bartz order, Judge Chhabria granted Meta summary judgment on fair use in Kadrey v. Meta Platforms, No. 3:23-cv-03417 (N.D. Cal.). Both judges found training on books transformative, and both ruled for the AI company on training. The reasoning diverged on two points. Judge Alsup treated the pirated library as a separate use that fair use could not excuse. Judge Chhabria treated Meta’s downloading as part of its training purpose and decided the case on market harm, holding that the authors had offered no meaningful evidence that Llama would dilute the market for their books (Kadrey order, Dkt. 598).

Question	Bartz v. Anthropic	Kadrey v. Meta
Training on books	Fair use, “exceedingly transformative”	Fair use on this record
Pirated downloads	Not fair use when kept as a library	Judged as part of the training purpose
Market dilution theory	Rejected as a copyright harm	Recognized in principle, but unproven here
Outcome	Class certified; $1.5 billion settlement	Meta won on fair use; torrenting claims continue

In June 2026 the Kadrey plaintiffs asked to certify the conflict for interlocutory appeal; Judge Chhabria denied that request on July 8, 2026, according to Mishcon de Reya’s case tracker. Neither ruling binds other judges, and the Ninth Circuit has not addressed the question as of November 9, 2026. A full comparison is on the Kadrey page in the related links.

What it means for authors and AI companies

For AI developers, the ruling separates two risks that had been discussed together. Training on lawfully acquired books was held to be fair use, at least where the plaintiffs did not allege infringing outputs. How the books were acquired is a separate exposure: a library assembled from pirate sites was infringing on its own, and statutory damages per work across hundreds of thousands of works made that exposure large enough to settle at $1.5 billion. That provenance theory has since been picked up by other plaintiffs. The music publishers’ second suit against Anthropic, filed in January 2026, cites piracy evidence surfaced in Bartz (TechCrunch).

For authors, the settlement pays only for books on the Works List and leaves other claims intact. Books that were not registered with the Copyright Office in time are outside the class, and output-based claims against Anthropic were not released. For the broader doctrine, see fair use in the age of AI; for the parallel news case, see New York Times v. OpenAI.

What happens next

The litigation is over: final judgment was entered on July 20, 2026. Class counsel’s September 2, 2026 status report says the only notices of appeal came from law firms excluded from the fee award and concern fees alone, and that the parties agree the settlement’s Effective Date has passed. What remains is administration. The parties’ October 23, 2026 stipulation expects the first distribution during the week of November 23, 2026. Disputes between authors and publishers over a work’s share go to a court-appointed Special Master, and the court retains jurisdiction over the settlement’s implementation.

Anthropic’s final $450 million installment is due by September 25, 2027, or earlier after a qualifying funding event, which will fund a second distribution. The court also withheld 10 percent of class counsel’s fee until a post-distribution accounting is filed. This page is general information, not legal advice.

What changed

• 2026-10-23: Parties set the initial settlement distribution for the week of November 23, 2026 (Dkt. 694).
• 2026-09-02: Class counsel reports the Effective Date has passed and estimates $2,203.56 per work for the first distribution (Dkt. 688).
• 2026-07-20: Final approval granted, judgment entered, case dismissed with prejudice (Dkt. 680).
• 2026-05-14: Fairness hearing before Judge Araceli Martinez-Olguin.
• 2026-01-05: Publishers Weekly reports the case has moved to Judge Martinez-Olguin after Judge Alsup's retirement.
• 2025-09-25: Preliminary approval of the $1.5 billion settlement.
• 2025-09-05: Settlement agreement filed.
• 2025-07-17: Class certified for owners of books downloaded from LibGen and PiLiMi.
• 2025-06-23: Order on fair use: training and print-to-digital scanning are fair use; pirated library copies are not (Dkt. 231).

Frequently asked questions

What was the ruling in Bartz v. Anthropic?

On June 23, 2025, Judge William Alsup held that training Claude on books was fair use and that scanning purchased print books was fair use, but that downloading and keeping pirated books in a central library was not. He set the piracy claims for trial before the case settled.

Is training AI on books fair use?

Two federal district judges in California have said yes on the records before them: Judge Alsup in Bartz and Judge Chhabria in Kadrey v. Meta. Neither ruling binds other courts, Chhabria stressed his was limited to the plaintiffs' evidence, and no appeals court has decided the question as of November 2026.

How much did Anthropic settle for?

Anthropic agreed to pay $1.5 billion plus interest into a non-reversionary fund, about $3,000 per work before fees and costs. Judge Araceli Martinez-Olguin granted final approval on July 20, 2026, and first payments are expected the week of November 23, 2026.

Does the Bartz ruling apply to other AI companies?

Not as binding precedent. A district court ruling does not bind other judges, and the case settled before any appeal. It has been persuasive, though: other plaintiffs, including music publishers suing Anthropic, have built claims around its distinction between training and pirated acquisition.

Sources

1. Order on fair use, Dkt. 231 (June 23, 2025)
2. Order granting final approval, Dkt. 680 (July 20, 2026)
3. Class counsel status report, Dkt. 688
4. Joint stipulation, Dkt. 694
5. Kadrey v. Meta order on fair use, Dkt. 598
6. Anthropic Copyright Settlement (official administrator site)
7. Authors Guild: What authors need to know
8. Authors Guild: Key updates on the settlement
9. Norton Rose Fulbright: settlement after summary judgment and class certification
10. Publishers Weekly: new judge to preside
11. Mishcon de Reya: generative AI IP case tracker
12. TechCrunch: music publishers sue Anthropic for $3B