top of page
Search

The $1.5 Billion Settlement Was Never Global

3 days ago
10 min read

What the Anthropic copyright story leaves out once you step outside the United States


Anthropic's $1.5 billion settlement with US authors made headlines as the largest copyright settlement on record. It's a landmark — but it's also a very American-shaped resolution to what is, in practice, a global problem. Step outside of US borders and the picture looks nothing like a tidy class-action payout. It looks like a registration technicality that quietly excludes most of the world's writers, a Japanese book distributor caught up in a contract dispute it may never have seen coming, and a German collecting society already winning the fight that no one has yet brought against Anthropic.


None of what follows is a lawsuite against Anthropic outside the US — as of this writing, there isn't one. But the settlement's edges are where the more interesting questions live.

The Registration Wall

The Bartz v. Anthropic settlement compensates a specific, narrow set of works: those found in the pirated datasets Anthropic used, that appear on the settlement's approved "Works List," and that were registered with the US Copyright Office — generally within five years of publication, or three months for the strongest protection. That last condition is where things get uneven fast.


Japanese investigative outlet SlowNews traced this detail: Haruki Murakami's Norwegian Wood, in English translation, turned up in Anthropic's training data via a Books3-style pirated dataset. Of the roughly 4 million unique works Anthropic downloaded, only 482,460 made the final, compensable list — and of an estimated 2.5 million non-English works in the dataset, the overwhelming majority were never registered with the US Copyright Office at all. Registration was never a meaningful option for most non-US authors in the first place; it isn't how copyright works in most of the world, where protection attaches automatically. The result is a settlement that pays out almost entirely along linguistic and jurisdictional lines that have nothing to do with whether a work was actually used.


Canada tells the same story from a different angle. The Writers' Union of Canada's guidance to its members is blunt: Canadian works can only participate if they meet the same US registration and dataset-inclusion criteria as anyone else. There's no separate Canadian claims process, no separate Canadian action — just an instruction to check the US portal and see if you got lucky.

The Domestic Wrinkle: What "Opting Out" Actually Means

Even for US authors who do fall within the settlement class, there's a second gate that's easy to get wrong — and it's the same gate Anthropic is now leaning on in its motion to dismiss the opt-out suits (the one we wrote up separately). The class was defined in the July 2025 certification order (Bartz v. Anthropic PBC, 791 F. Supp. 3d 1038 (ND Cal 2025)) to cover owners of the reproduction right in books found in the LibGen/PiLiMi libraries Anthropic downloaded, subject to registration requirements. The settlement itself moved from a term sheet (August 25, 2025) to a signed agreement (September 5, 2025) to preliminary court approval (October 17, 2025, Bartz v. Anthropic PBC, 2025 US Dist. LEXIS 205531) to final approval on July 20, 2026 — nearly a year after the case was filed on August 19, 2024.


Two dates that matter for anyone still working through their own position, both now confirmed against the court's own final order rather than estimated:

  • The opt-out/objection deadline was February 9, 2026 — the final date, after two extensions (originally January 7, then pushed to January 29 following the reassignment from Judge Alsup to Judge Araceli Martinez-Olguin, the to February 9)

  • The release cutoff is August 25, 2025 — the settlement notice is explicit that "Released Claims do not extend to any activity or conduct that occurs or occured after August 25, 2025." It covers Anthropic's downloading, copying, and training conduct through that date; it does not release claims about AI model outputs or conduct after it.


Refusing the settlement check is not the same thing as preserving your right to sue, and the distinction trips people up constantly:

  • Opt out by the deadline (February 9, 2026, as it turned out), and you keep your right to sue Anthropic individually — but you're still racing the ordinary three-year copyright statute of limitations (17 USC § 507), which runs from when you discovered the infringement, not from any settlement date. Filing the class action can pause that clock while you're a class member (Crown, Cork & Seal Co. v. Parker, 462 US 345 (1983)). If you discovered the use well before the class action was filed on August 19, 2024, you may have opted out with very little runway left.

  • Stay in the class and simply not file a claim, and courts have consistently held you're still bound by the release — whether or not you ever saw a dollar. (Richardson v. Wells Fargo Ban, N.A., 839 F.3d 442 (9th Cir. 2016)). Decling to cash a check doesn't undo that.

  • Objecting without opting out, and the same is true — an objection tells the court you think the deal is unfair, but it doesn't remove you from the class (Phillips Petroleum Co. v. Shutts, 472 US 797 (1985)).


That's the legal mechanism underneath the motion to dismiss: Anthropic isn't arguing the opt-out authors lack a claim on the merits — it's arguing the clock already ran before they filed. Given that tolling pauses rather than resets the limitations period, that's a genuine risk for anyone who says they knew about the alleged use well before mid-2024 — and with only 350 opt-outs on record, it's a small, identifiable group. None of this is specific to Bartz, either — it's the general framework (Warner Chappell Music, Inc. v. Nealy, 601 US 366 (2024)), on discovery-based accrual; Concord Music Grp., Inc. v. Anthropic PBC, 2025 US Dist. LEXIS 279774, on a related DMCA claim against Anthropic) that will govern the next mass AI-copyright settlement too, whenever it comes.

When Your Own Distributor is the Counterparty

What started as speculative legal commentary is now a confirmed, multi-outlet story with a name, a contract date, and a corporate denial on the record. Nikkei Business broke it on September 4, 2026: court filings in Bartz v. Anthropic (ND Cal., No. 3:24-cv-05417) — an expert witness declaration and Anthropic's own internal finance-meeting minutes — name Nippan (Nippon Shuppan Hanbai), Japan's largest book distributor, as a supplier. The filings put contract negotiations beginning September 18, 2024, and the contract executed November 21, 2024. Anthropic reportedly built a dedicated internal tool — nicknamed the "Nippan Loader" — to convert Japanese book-classification codes into its own systems, part of a broader program it called "Project Panama": buying physical books and destructively scanning them (cutting the spine, scanning each page, discarding the original), with one internal note from May 2024 reportedly saying the plan was to "store everything forever." Quantities and per-unit pricing remain redacted.


Asahi Shimbun followed up on September 18 with Nippan's on-record response: the company said there is "no fact of selling with recognition it was for AI company learning purposes." MLEX picked the story up in English on September 29 ("Anthropic's US book case spills into Japan's publishing trade," by Toko Sekiguchi), reporting that Japanese publishers are now asking whether their own distribution agreements with Nippan ever authorized this in the first place. The Japan Book Publishers Association (JBPA — 381 member publishers) has sent Nippan a formal letter requesting confirmation of what happened.


The legal analysis underneath the headlines — laid out by Japanese IP counsel writing on the disclosure — is why this is a harder problem for publishers than it looks. Japan's first-sale doctrine (copyright law, Article 26-2) means that once a book is sold into commerce, reselling it isn't a copyright violation — publishers can't stop a distributor from selling books onward on copyright grounds alone. Standard Japanese distribution and resale-price contracts, written before generative AI existed, typically say nothing about overseas resale or AI training as an end user. That leaves two possible levers, and they're both weaker than a copyright claim: explicit contract restrictions on transfer or purpose (which mostly don't exist yet), or Japan's Article 30-4 "information analysis" exception — a TDM-style carveout that permits use for AI training unless it would unreasonably prejudice the rightsholder's market, a standard nobody has tested against a case like this one.


In other words: even in a jurisdiction, like Japan, with more AI-permissive copyright regime than the US, publishers have less to work with than their American counterparts did — because the leverage in Bartz came from proving the acquisition was piracy, and the Nippan sale, on the available facts, looks like a legitimate commercial transaction. (Worth noting, for anyone modeling motive: Nippan's core distribution division posted a ¥3.6 billion operating loss on ¥298.2 billion in sales, down 12.3%, in its most recent fiscal year — a large, no-returns order from a single buyer is exactly the kind of revenue a shrinking distributor doesn't say no to.)

A Second Thread — Related, But Not the Same Story

Separtely, starting around August 2026, Japanese used bookstores reported a five-fold surge in bulk online orders from unidentified buyers, with an estimated 50 tons of books — roughly 100,000 volumes — consolidating at a logistics center in Okayama Prefecture before export to the US (Cybernews, Tom's Hardware, Eastern Herald). Nippon TV (NNN) initially reported this alongside the Nippan story as though it were the same operation. It probably isn't, or at least isn't provably so: a closer read of the court filings (credited to Japanese AI commentator Tanu) found no document tying the 50-ton used-book export specifically to Anthropic — only that the timing loosely lines up with Nippan's new classification system. Worth keeping the two threads separate: one is a named, dated contract now sitting in a US federal court record; the other is an unattributed demand spike that several outlets have connected to "AI training" in general, without confirming a buyer.

Germany is Playing a Different Game

If there's a template for what real, non-US litigation against an AI company looks like, it's Germany's GEMA — the country's performing rights society — which has already won cases this year against OpenAI and against the AI music generator Suno over unlicensed use of song lyrics, litigated in German courts under German law. GEMA hasn't sued Anthropic yet. But the theory is directly transferable, and collecting societies exist precisely to bring the kind of coordinated action that individual authors can't. If a German case against Anthropic materializes, GEMA's playbook against OpenAI and Suno is the one to watch.


France, by contrast, has produced commentary rather than a case of its own: French coverage of the Concord Music Group v. Anthropic lyrics litigation (still a US case, in the Northern District of California) has focused on the amicus briefs filed in March 2026 by international publishers and author associations — a sign of concern, not a French cause of action.

The Practitioner Takeaway

For anyone advising clients with cross-border rights exposure, the lesson isn't "the settlement covers you" or "the settlement doesn't cover you" — it's that the US class-action model simply doesn't export. Each jurisdiction is going to answer the "was our work used, and do we have recourse" question with its own doctrinal toolkit: US formalities (registration) gate the class settlement; Japanese first-sale and contract law gate the Nippan situation, and now that there's a dated, named contract on the record, the live question for Japanese publishers is exactly the one JBPA is asking Nippan — what did the distribution agreement actually permit; Germany's collecting-society infrastructure is what makes GEMA-style case possible at all; France, for now, is watching from the sidelines.


If you represent a non-US author, publisher, or distributor with any exposure to AI training pipelines, the actionable question isn't "am I in the settlement" — it's "what does my contract say," and "does my jurisdiction have a collecting society or statutory exception built for exactly this fight." Those answers are going to look very different depending on where your client sits.

Not sure whether your rights, your distribution agreements, or your content are exposed to an AI training pipeline — especially if you or your rightsholders sit outside the US?

We help authors, publishers, and distributors sort through exactly this kind of cross-border ambiguity: contract and distribution-agreement review, licensing strategy, and plain-English guidance on where the law actually stands in your jurisdiction today. Book a free consultation call with us and we'll help you figure out where you stand.

📅 Book online: www.atmoip.com/book-online

🌐 Visit: www.atmoip.com

📩 Email: info@atmoip.com


Have a take on where this is headed? Drop it in the comments — we read every one.

You can also follow aTMospheric IP (@atmoip) for practical insights and updates.




Wherever your rights sit, they're worth defending.

 

stylized wording aTMosperic IP

HQ 371 NE GILMAN BLVD., STE 160

1ST FLOOR

ISSAQUAH, WA 98027

Sources:

On the registration wall / Japan & Canada exclusion from the settlement:


On the opt-out / statute-of-limitations mechanics (from legal research):

  • Bartz v. Anthropic PBC, 787 F. Supp. 3d 1007 (N.D. Cal. 2025) — fair-use ruling

  • Bartz v. Anthropic PBC, 791 F. Supp. 3d 1038 (N.D. Cal. 2025) — class certification

  • Bartz v. Anthropic PBC, 2025 U.S. Dist. LEXIS 205531 (N.D. Cal. Oct. 17, 2025) — preliminary settlement approval

  • Concord Music Grp., Inc. v. Anthropic PBC, 2025 U.S. Dist. LEXIS 279774 (N.D. Cal. 2025)

  • Warner Chappell Music, Inc. v. Nealy, 601 U.S. 366 (2024)

  • Crown, Cork & Seal Co. v. Parker, 462 U.S. 345 (1983)

  • China Agritech, Inc. v. Resh, 584 U.S. 732 (2018)

  • Richardson v. Wells Fargo Bank, N.A., 839 F.3d 442 (9th Cir. 2016)

  • Phillips Petroleum Co. v. Shutts, 472 U.S. 797 (1985)


On the confirmed opt-out deadline, release cutoff, and final approval (primary sources):


On the Nippan contract:

Referenced but paywalled — findable via search, unable independently verify in full:

  • Nikkei Business, Sept. 4, 2026 (broke the Nippan story)

  • Asahi Shimbun via Yahoo! News Japan, Sept. 18, 2026 (Nippan's on-record denial)


On the separate ~50-ton used-book story:


On Germany and France:


A note on sourcing: most of the Nippan details above come from Japanese-language commentary and investigative newsletters (such as SlowNews, note.com writers), they were read and summarized in translation rather than by a native reader.


Disclaimer: This Blog post and any accompanying materials are for informational purposes only and are not intended to provide, nor should they be relied upon as, legal advice. Legal rules can change often. The information contained herein is accurate to the best of our knowledge as of the date of this post. 

Comments

© 2024-2026 by aTMospheric IP, LLC. All rights reserved.

bottom of page