Table of Contents
- What Bartz v. Anthropic Actually Is
- What’s Happening on May 14
- How the Settlement Pays Out
- The Five Objections the Judge Has to Rule On
- Why This Might Still Not Get Approved
- What Happens After the Hearing
- What This Means for the Rest of the AI Industry
- FAQ
On May 14 at 2 p.m. Pacific, in Courtroom 12 of the San Francisco Federal Courthouse, a judge will decide whether to grant final approval to the largest copyright settlement in U.S. history. The settlement is $1.5 billion. The case is Bartz v. Anthropic. The class includes 482,460 books — most of them downloaded by Anthropic from pirate libraries to train Claude. As of the latest filing, 440,490 of those works (91.3%) have been claimed. Each one is set to pay out roughly $3,000 to the rightsholder.
The headline number is real. The question on May 14 is not whether the settlement is large enough to make news. It’s whether five categories of objection from authors and literary organizations are serious enough to make the judge reject the deal as written.
What Bartz v. Anthropic Actually Is
The lawsuit was filed in August 2024 by three authors — Andrea Bartz, Charles Graeber, and Kirk Wallace Johnson — alleging that Anthropic had downloaded millions of copyrighted books from shadow libraries including Library Genesis and Pirate Library Mirror, then used those books as training data for Claude.
The case was a class action from the start. Discovery confirmed Anthropic had ingested at least 7 million pirated books, including works by every major American publisher and tens of thousands of independent authors. Anthropic’s defense centered on fair use for training purposes. That defense did not survive the summer of 2025: in August, Anthropic agreed to a $1.5 billion settlement to resolve the class claims before trial. Susman Godfrey, the plaintiffs’ counsel, took the case from filing to settlement in under a year.
The class period ran from January 2020 to August 2025. Eligible works are books with active U.S. copyright registration that appeared in Anthropic’s training corpus. Both authors and publishers can claim, with payment splits based on the underlying contracts.
What’s Happening on May 14
The May 14 hearing is the final approval (sometimes called the fairness) hearing. Federal class action settlements require court approval at two stages: preliminary, which happened in late 2025, and final, which happens after the claim window closes and objectors have had time to file.
The judge will consider three things at the hearing. First, whether the settlement is fair, reasonable, and adequate to the class as a whole — the legal standard under Rule 23. Second, whether class notice was sufficient and the claim rate signals real participation. Third, whether the objections raised by individual class members or amici are serious enough to require renegotiation.
Class members who filed timely objections may attend in person or via Zoom and address the court directly. Several have indicated they will. The hearing is at 450 Golden Gate Avenue, Courtroom 12 on the 19th floor. It is open to the public.
How the Settlement Pays Out
The $1.5 billion fund is distributed as follows. First, attorneys’ fees and costs come off the top — class counsel is requesting roughly 25% of the fund, which would be approximately $375 million. Administration fees for the claims process come off next. The remainder — somewhere in the range of $1.05 to $1.1 billion depending on final fee approval — goes to claimants.
Per-work compensation is approximately $3,000 before pro rata adjustment. With 91.3% of eligible works claimed, the per-work payment may end up close to the $3,000 figure rather than being diluted significantly higher. Splits between authors and publishers depend on individual publishing contracts; works with reverted rights pay 100% to the author, while works with active publisher rights typically split based on the contract’s royalty structure.
Eligible class members can still claim if they file before the final cutoff, though late claims may be subject to additional review. The settlement website at anthropiccopyrightsettlement.com remains open for verification and claim status.
The Five Objections the Judge Has to Rule On
Objections fall into five categories, and the judge has to address each before granting final approval.
1. Exclusion of foreign and non-US-registered works. The class is limited to works with active U.S. copyright registration. Foreign authors whose books Anthropic ingested without U.S. registration are excluded. Objectors argue this carves out a meaningful fraction of harmed authors and treats their works as if they don’t count.
2. Systematic publisher favoring over authors. Several author advocacy groups argue that the per-work split mechanism — driven by individual publishing contracts — systematically favors traditional publishers over the authors whose works were actually ingested. The objection is structural, not arithmetic: the settlement’s distribution architecture makes the average author’s check smaller than the average publisher’s, even though the harm was to authorship.
3. Misleading class notice. Some objectors argue the notice sent to class members underplayed both the scope of the harm and the alternative of opting out to pursue individual claims. The argument is that authors who would have opted out had they understood the full picture instead defaulted into the settlement.
4. Inadequate compensation relative to damages. Statutory damages for willful copyright infringement run up to $150,000 per work. The settlement pays roughly $3,000 per work — a 50:1 discount on the statutory ceiling. Objectors note that Anthropic’s own training data acquisition was, by the company’s stipulated facts, a bulk download from pirate libraries; that’s the textbook fact pattern for willful infringement, and the discount accordingly should be smaller.
5. Class counsel conflicts. Several objectors have raised concerns about Susman Godfrey’s representation, including the size of the requested fee award and the speed of settlement negotiation. The argument is that a faster settlement was more lucrative for class counsel than a slower trial would have been for class members.
Why This Might Still Not Get Approved
Final approval at fairness hearings is the default outcome — judges usually approve the deal in front of them rather than blow it up — but it isn’t guaranteed. The two paths to disapproval are: the judge agrees with one or more objections and orders renegotiation, or the judge approves with conditions that materially change the distribution.
The most likely friction points are objections #1 (foreign works) and #2 (publisher-favoring distribution). Either could plausibly produce a conditional approval that requires modifications before the fund actually pays out. Objection #4 (statutory damage discount) is harder to land — judges usually accept that settlements involve discounts, and a 50:1 ratio, while high, is within precedent for class action AI cases.
The longer the hearing runs, the more likely conditional approval becomes. A short hearing typically signals the judge is satisfied with the deal as written.
What Happens After the Hearing
If the judge grants unconditional final approval, the settlement becomes effective immediately. Distribution begins after a 30-day appeal window. Most class members should receive payment within 90 to 120 days of final approval.
If the judge grants conditional approval — for instance, requiring revised distribution or a modified opt-out window for excluded foreign works — the settlement returns to the parties for renegotiation. Distribution is delayed until the conditions are met. This is the most likely path if the objections gain traction.
If the judge denies approval outright, the case returns to active litigation. That outcome is unlikely but not impossible, and would push any resolution at least 12 to 18 months out. It would also reset Anthropic’s legal exposure to the full statutory damages range.
What This Means for the Rest of the AI Industry
Bartz v. Anthropic is the first major copyright settlement in the AI training data era. Whatever happens on May 14 sets the price floor for every other case still in litigation: OpenAI faces parallel actions from the New York Times and other publishers, Microsoft is named in some of them, and several music publisher cases against major AI labs are in active discovery.
The $3,000-per-work number, even before pro rata adjustment, is now the reference point. Plaintiffs in other cases will use it to argue for similar or higher settlements. Defendants will use it to argue against significantly larger ones. The objection categories that survive May 14 — particularly any judicial findings on publisher-versus-author distribution or on excluded foreign works — become precedent for how every subsequent settlement is structured.
This is also the first time a U.S. court has implicitly priced the harm of training on copyrighted works without permission. The number is not zero. The number is also not the full statutory ceiling. That middle ground — $3,000 per work, sourced from a $1.5 billion fund — is now the reference rate for the entire industry.
FAQ
Am I eligible if I’m an author whose book was used to train Claude?
Eligibility requires the work to have active U.S. copyright registration as of the class period (January 2020 through August 2025) and to have appeared in Anthropic’s training corpus. The settlement website at anthropiccopyrightsettlement.com has a search tool to check work-by-work. Foreign authors whose works do not have U.S. registration are excluded under the current settlement terms.
How much will I actually receive per book?
Approximately $3,000 per work before pro rata adjustment, less applicable attorneys’ fees and administration costs. Final per-work payment depends on the total claim rate and approved fee award. With 91.3% claimed, dilution upward is limited but possible.
What if my book is published by a traditional publisher?
Publisher-author splits depend on the underlying publishing contract. Works with reverted rights typically pay 100% to the author. Works with active publisher rights pay according to the contract’s royalty structure, which often favors the publisher significantly. This split mechanism is one of the live objections at the May 14 hearing.
Can I opt out and sue Anthropic individually?
The opt-out window has closed for current class members. If the judge grants final approval, the settlement releases all class claims and individual lawsuits become foreclosed for participating members. If the judge orders a modified settlement with a new opt-out window, that creates a fresh option to exit and pursue individual claims.
Does this settle Anthropic’s other copyright exposure?
No. Bartz covers the specific class of U.S.-registered books in Anthropic’s training corpus during the class period. Other Anthropic legal matters, future training data acquisitions, and non-book copyrighted works (music, code, journalism, images) remain separate exposures with their own litigation paths.
What does this settlement do to AI training going forward?
It establishes a price tag for the unauthorized use of copyrighted works in training corpora. Going forward, AI labs that have ingested similar data face a known dollar-figure precedent. It also accelerates licensing-deal pressure: major model providers are increasingly negotiating data licensing agreements proactively rather than relying on the fair-use defense that did not survive the Bartz pleadings.