Anthropic and the Pirate Library: The $1.5 Billion Copyright Bill
There are cheap ways to build a library.
There are expensive ways to build a library.
And then there is the Anthropic way: download millions of books from pirate websites, feed them into the machinery of artificial intelligence, get sued by authors — and eventually agree to a $1.5 billion settlement.
Welcome to the new economics of copyright.
The remarkable thing about Bartz v. Anthropic is not simply the size of the bill. It is what the court said before the bill arrived.
Anthropic won one of the most important arguments in the entire AI copyright war.
And then it walked straight into another one.
The Library Behind Claude
Anthropic is the company behind Claude, one of the world's major generative AI systems.
Like other large language models, Claude had to learn from text. An enormous amount of text.
Books are particularly attractive training material. They contain carefully edited language, long-form reasoning, narrative structure, specialist knowledge and millions upon millions of examples of humans doing that peculiar human activity known as writing.
There was only one irritating detail.
Books have owners.
According to the litigation, Anthropic accumulated millions of books while assembling its research and training library. Some were eventually obtained legitimately. But millions of digital books had also been downloaded from notorious shadow libraries, including Library Genesis (LibGen) and Pirate Library Mirror (PiLiMi).
These were not mysterious fragments drifting around the public internet.
They were books.
Whole books.
Written by people who, inconveniently for the economics of artificial intelligence, possessed copyrights.
Authors Andrea Bartz, Charles Graeber and Kirk Wallace Johnson sued Anthropic in 2024.
And that is where the story became much more interesting than another routine tale of internet piracy.
Reading Is Not Stealing. Apparently Neither Is AI Training.
In June 2025, U.S. District Judge William Alsup handed Anthropic an extraordinary victory.
Using copyrighted books to train its AI models could constitute fair use.
The reasoning was enormously important.
Training an AI system was treated as transformative rather than simply another method of distributing copies of the original books. The model was not a bookstore selling replacement copies of novels. It was learning statistical relationships from enormous quantities of text in order to produce new outputs.
Judge Alsup compared the process, in substance, to humans learning by reading.
That was a potentially historic victory for the AI industry.
If every copyrighted work required an individual training license before a machine could learn from it, building modern large language models could become legally and economically nightmarish.
Anthropic had therefore won a very large battle.
Unfortunately, it had another problem.
Where did the books come from?
Fair Use Is Not a Receipt
Imagine buying a book.
You read it. You study it. You learn its structure. Perhaps you become a writer yourself.
Copyright law does not normally require you to erase everything you learned.
Now imagine that instead of buying the book, you walk out of the store with it hidden under your coat.
Your subsequent literary education does not magically transform the shoplifting into a lawful transaction.
That distinction became the brutal center of Bartz v. Anthropic.
Judge Alsup separated two acts that much of the public debate had casually thrown into the same bucket:
using a work to train an AI
and
obtaining an unauthorized copy of that work in the first place.
Anthropic could have a fair-use argument for the first.
That did not automatically give it one for the second.
The court found that Anthropic's creation of a permanent central library containing millions of pirated books was not protected merely because some of those books might later be useful for AI training.
That distinction is devastatingly simple.
You may have the right to use something in a particular way.
That does not necessarily give you the right to steal the copy you use.
Then Came the Calculator
Copyright litigation becomes especially entertaining when lawyers stop discussing philosophy and open a calculator.
Under U.S. copyright law, statutory damages for willful infringement can reach $150,000 per work.
Now multiply unpleasant numbers by hundreds of thousands of books.
Suddenly $1.5 billion starts looking less like an astronomical punishment and more like an escape hatch.
The certified settlement ultimately covered roughly 482,000 works, with approximately $3,000 allocated per eligible title before applicable deductions and allocation among rightsholders.
Anthropic agreed to the $1.5 billion settlement in 2025.
On July 20, 2026, the federal court gave it final approval. More than 91% of eligible works had claims filed, according to reporting on the final approval.
The settlement also required the destruction of the pirated copies covered by the agreement.
One and a half billion dollars.
For books that could, in many cases, have been purchased legally for twenty dollars.
There is something almost artistically Silicon Valley about the mathematics.
The Strangest Part: Anthropic Was Also Buying Books
And now the story becomes wonderfully absurd.
Anthropic did not rely exclusively on pirate libraries.
The company also acquired physical books legitimately and digitized them for its internal library. The court treated the conversion of lawfully purchased books into digital form for this purpose very differently from downloading unauthorized copies from pirate repositories.
So we arrive at a copyright distinction that future historians may enjoy immensely:
Buy the book, digitize it and use the resulting material for transformative AI training?
Potentially fair use.
Download the same book from a pirate library and keep the unauthorized copy in a giant corporate archive?
Potential copyright infringement.
Same words.
Same neural network.
Same training process.
Completely different legal problem.
Copyright, as usual, cares very much about how you got into the room.
The $1.5 Billion Question
The decision did not establish the simple proposition that:
AI companies must pay authors whenever copyrighted books are used for training.
In fact, one of the most significant parts of the ruling pointed in almost the opposite direction.
Training on lawfully acquired books could qualify as fair use.
That matters enormously.
The $1.5 billion settlement therefore should not be casually described as "$1.5 billion for training Claude on copyrighted books."
That misses the interesting part.
The legal bomb was the pirated library.
Anthropic's problem was not merely that its machine had read copyrighted books.
Its problem was that somebody had first assembled a mountain of unauthorized copies.
The difference sounds technical until someone attaches a billion-dollar price tag to it.
Then everybody suddenly understands.
Copyright Meets the Industrial Age of Copying
Piracy used to have a wonderfully modest scale.
Someone photocopied a textbook.
Someone copied a cassette.
Someone downloaded Harry Potter from a torrent site because apparently civilization would collapse if they paid $9.99.
Generative AI changed the scale.
A modern AI company does not necessarily want one unauthorized book.
It may want millions.
Not because it intends to sell those books individually, but because the collection itself becomes industrial raw material.
That creates a copyright problem that existing vocabulary handles rather badly.
The AI company says:
We aren't competing with the book. We're learning from it.
The author says:
Wonderful. Why did you need a pirate copy of my book to do that?
Both statements can be true.
That is precisely why Bartz v. Anthropic matters.
Nobody Got the Clean Victory They Wanted
Authors did not receive a judicial declaration that AI training on copyrighted books is inherently illegal.
Quite the contrary.
Anthropic obtained an extremely important fair-use ruling.
But AI companies did not receive a universal judicial permission slip saying that transformative training sanitizes whatever happened upstream.
It doesn't.
A transformative purpose is not holy water.
You cannot necessarily pour it over an infringing copy and watch the infringement disappear.
The case therefore produced something far more useful than a simplistic winner and loser.
It produced a boundary.
Training may be fair use. Piracy may still be piracy.
That boundary will matter far beyond Anthropic.
AI developers need gigantic datasets. Publishers possess gigantic catalogs. Authors possess copyrights. Shadow libraries possess gigantic quantities of everybody else's books.
And now everyone has a rather spectacular number to contemplate:
$1,500,000,000.
The New Cost of “Download Everything”
For years, the culture surrounding large-scale data collection operated on an almost adolescent principle:
Take first.
Build something enormous.
Ask the lawyers later.
The internet encouraged it. Storage became cheap. Bandwidth became cheap. Scraping became easy. Venture capital became abundant.
Copyright did not disappear.
It merely became inconvenient.
Bartz v. Anthropic is what happens when that inconvenience finally receives an invoice.
The lesson is not that artificial intelligence must stop reading.
The lesson may be much simpler.
If your billion-dollar machine needs a library—
perhaps buy the damn books.
Case: Bartz et al. v. Anthropic PBC
Court: U.S. District Court, Northern District of California
Filed: 2024
Key ruling: June 2025
Settlement: $1.5 billion
Final approval: July 20, 2026
Subject: Copyright infringement, generative AI training, fair use, pirated books
Copyright Registration will cost - $7
Public offer - read carefully before registration!
SciReg Terms of Service
1. General Provisions
These Terms of Service ("Terms") constitute a legally binding agreement between you ("Customer," "User," or "you") and SolCity Nav, LLC, a Delaware limited liability company ("SciReg," "Company," "we," "our," or "us"), governing your access to and use of the SciReg website and all related services.
By creating an Account, submitting Registration(s), uploading Submitted Materials, purchasing Registration(s), or otherwise using the Service, you acknowledge that you have read, understood, and agree to be bound by these Terms.
If you do not agree to these Terms, you must not access or use the Service.
The Company reserves the right to modify these Terms at any time. Revised Terms become effective immediately upon publication on the SciReg website. Your continued use of the Service after publication of revised Terms constitutes your acceptance of those revisions.
2. Definitions
For purposes of these Terms:
Account means a registered user account created through the SciReg website.
Customer means any individual or legal entity using the Service.
Service means the SciReg website and all services provided by SolCity Nav, LLC.
Registration means the submission of information and Submitted Materials through the Service for creation of a Registration Record.
Registration Record means the electronic record created and maintained by SciReg containing information supplied by the Customer together with the assigned registration number and registration date.
Registration Certificate means the electronic certificate generated by SciReg confirming creation of a Registration Record.
Submitted Materials means any files, documents, manuscripts, publications, software, databases, images, audiovisual works, research materials, or other content submitted by the Customer through the Service.
3. Description of the Service
SciReg is an independent online registration and publication service that enables Customers to create dated electronic Registration Records for intellectual works and other Submitted Materials.
For each accepted Registration, SciReg may:
- assign a unique registration number;
- record the date and time of Registration;
- maintain a Registration Record;
- securely store Submitted Materials;
- apply a digital signature to the Submitted Materials to enable verification of their integrity and detection of any subsequent modifications;
- generate and issue a Registration Certificate;
- apply a digital signature to the Registration Certificate to enable verification of its authenticity and integrity;
- publish Registration information where publication is selected or otherwise authorized by the Customer.
SciReg operates as a private commercial service and is not affiliated with any governmental authority.
4. Nature of the Service
SciReg is an independent private registration service.
SciReg is not:
- a government copyright office;
- a patent office;
- a trademark office;
- a governmental registration authority;
- a court of law;
- a notary public;
- an arbitration tribunal.
SciReg does not:
- determine authorship;
- determine ownership of intellectual property rights;
- verify the identity of Customers;
- verify the accuracy of submitted information;
- verify originality of Submitted Materials;
- determine whether Submitted Materials qualify for copyright or other legal protection under any jurisdiction;
- provide legal advice or legal opinions;
- resolve disputes concerning ownership or intellectual property rights.
SciReg creates Registration Records solely on the basis of information provided by the Customer. The Company does not independently verify the identity of the Customer or the accuracy, completeness, legality, ownership, originality, authenticity, or legal status of any Submitted Materials.
A Registration Certificate confirms only that SciReg received specified information and Submitted Materials, created a Registration Record, assigned a registration number, and recorded the corresponding date and time of Registration.
A Registration Certificate is not a governmental registration, is not evidence of ownership or authorship, does not establish the existence or validity of copyright or any other intellectual property rights, and does not constitute a legal opinion, certification, adjudication, or governmental act.
5. Registration Procedure
To create a Registration Record, the Customer must:
- Create a SciReg Account or sign in to an existing Account.
- Complete the Registration form by providing the required information.
- Upload the Submitted Materials. Each Registration may include one archive file not exceeding 7 MB in size.
- Pay the applicable Registration fee, unless the Registration qualifies for a free Registration offered by the Company.
- Submit the Registration through the Service.
Upon successful completion of the Registration process and receipt of payment, if applicable, SciReg will create a Registration Record, assign a unique registration number, record the date and time of Registration, and make a Registration Certificate available to the Customer.
As part of a completed Registration, SciReg applies a digital signature to the Submitted Materials and the Registration Certificate. The digital signature is intended to verify the integrity of the signed files, confirm that the Registration Certificate originated from SciReg, and detect any subsequent modifications to the signed files.
The Company reserves the right to reject or cancel any Registration that violates these Terms or applicable law.
6. Registration Fees
The Company may offer free Registrations and paid Registrations.
Unless otherwise stated on the SciReg website:
- the first two (2) Registrations submitted by a Customer are provided free of charge;
- each additional Registration is subject to a fee of US $7.00.
Registration fees are subject to change at any time. Any revised fees apply only to Registrations submitted after publication of the updated pricing.
All payments are non-refundable except where required by applicable law or expressly approved by the Company in its sole discretion.
Taxes, duties, bank charges, currency conversion fees, and similar charges imposed by third parties remain the sole responsibility of the Customer.
7. Customer Responsibilities
The Customer represents and warrants that:
- the Customer has the legal right to submit the Submitted Materials;
- the Submitted Materials do not knowingly violate any applicable law;
- the Submitted Materials do not knowingly infringe the intellectual property or other legal rights of any third party;
- all information provided during Registration is accurate and complete to the best of the Customer's knowledge.
The Customer is solely responsible for the content, legality, accuracy, ownership, and use of all Submitted Materials.
If Submitted Materials contain confidential, proprietary, trade secret, personal, or otherwise sensitive information, the Customer is solely responsible for determining whether submission to the Service is appropriate.
The Customer remains solely responsible for determining whether any AI-generated or AI-assisted materials qualify for copyright or other legal protection under applicable law.
8. License Granted to SciReg
The Customer retains all ownership and intellectual property rights in the Submitted Materials.
By submitting materials through the Service, the Customer grants SciReg a non-exclusive, worldwide, royalty-free license to:
- receive and process the Submitted Materials;
- create and maintain Registration Records;
- store backup copies;
- generate Registration Certificates;
- reproduce Submitted Materials as reasonably necessary for operation of the Service;
- publish Registration information where publication is selected or authorized by the Customer.
This license exists solely for the purpose of operating the Service and terminates when continued retention is no longer required for operation of the Service, legal compliance, dispute resolution, or legitimate business purposes.
Nothing in these Terms transfers ownership of the Submitted Materials to the Company.
9. Prohibited Use
The Customer agrees not to use the Service:
- for unlawful purposes;
- to submit malicious software or harmful code;
- to upload material that intentionally infringes the rights of third parties;
- to interfere with operation or security of the Service;
- to attempt unauthorized access to Company systems or other user Accounts;
- to submit false or misleading Registration information.
The Company reserves the right to refuse, suspend, remove, or terminate any Registration or Account that violates these Terms or applicable law without prior notice.
10. Privacy
The Company collects, stores, and processes personal information solely for the purpose of operating the Service, maintaining Registration Records, communicating with Customers, processing payments, complying with legal obligations, preventing fraud, and improving the Service.
The Company does not sell Customers' personal information to third parties.
By using the Service, the Customer consents to such processing of personal information.
11. Electronic Communications
The Customer agrees that all notices, disclosures, Registration Certificates, invoices, receipts, and other communications provided electronically by the Company satisfy any legal requirement that such communications be in writing.
12. Intellectual Property
The SciReg website, software, design, logos, trademarks, text, graphics, databases, and all other materials provided by the Company are the exclusive property of SolCity Nav, LLC or its licensors and are protected by applicable intellectual property laws.
Except as expressly permitted by these Terms, no part of the Service may be copied, reproduced, modified, distributed, reverse engineered, or otherwise used without the Company's prior written permission.
13. Availability of the Service
The Company will use commercially reasonable efforts to keep the Service available.
However, SciReg does not guarantee uninterrupted operation or continuous availability.
The Service may be modified, suspended, restricted, or discontinued at any time without prior notice.
The Company may perform maintenance, upgrades, security improvements, or other technical operations that temporarily interrupt the availability of the Service.
14. Disclaimer of Warranties
THE SERVICE IS PROVIDED "AS IS" AND "AS AVAILABLE."
TO THE MAXIMUM EXTENT PERMITTED BY APPLICABLE LAW, THE COMPANY DISCLAIMS ALL WARRANTIES, WHETHER EXPRESS, IMPLIED, STATUTORY, OR OTHERWISE, INCLUDING, WITHOUT LIMITATION, ANY IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE, NON-INFRINGEMENT, ACCURACY, RELIABILITY, OR AVAILABILITY.
THE COMPANY DOES NOT WARRANT THAT:
- THE SERVICE WILL OPERATE WITHOUT INTERRUPTION;
- THE SERVICE WILL BE ERROR-FREE;
- DEFECTS WILL BE CORRECTED;
- THE SERVICE WILL ALWAYS BE AVAILABLE;
- THE SERVICE WILL MEET THE CUSTOMER'S EXPECTATIONS OR PARTICULAR REQUIREMENTS.
15. Limitation of Liability
TO THE MAXIMUM EXTENT PERMITTED BY LAW, SOLCITY NAV, LLC, ITS OWNERS, DIRECTORS, OFFICERS, EMPLOYEES, CONTRACTORS, AGENTS, AND AFFILIATES SHALL NOT BE LIABLE FOR ANY INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, CONSEQUENTIAL, OR PUNITIVE DAMAGES, INCLUDING WITHOUT LIMITATION LOSS OF PROFITS, LOSS OF DATA, LOSS OF BUSINESS, LOSS OF GOODWILL, OR INTERRUPTION OF BUSINESS ARISING OUT OF OR RELATING TO THE USE OF THE SERVICE.
THE COMPANY SHALL NOT BE LIABLE FOR ANY LOSS OR DAMAGE RESULTING FROM:
- INFORMATION PROVIDED BY CUSTOMERS;
- UNAUTHORIZED ACCESS TO CUSTOMER ACCOUNTS;
- THIRD-PARTY ACTIONS;
- INTERNET FAILURES;
- FORCE MAJEURE EVENTS;
- VIRUSES OR OTHER HARMFUL SOFTWARE NOT CAUSED BY THE COMPANY'S GROSS NEGLIGENCE OR WILLFUL MISCONDUCT.
TO THE MAXIMUM EXTENT PERMITTED BY LAW, THE TOTAL LIABILITY OF THE COMPANY ARISING OUT OF ANY CLAIM RELATING TO THE SERVICE SHALL NOT EXCEED THE TOTAL AMOUNT PAID BY THE CUSTOMER FOR THE REGISTRATION GIVING RISE TO THE CLAIM.
16. Indemnification
The Customer agrees to defend, indemnify, and hold harmless SolCity Nav, LLC, its owners, directors, officers, employees, contractors, and affiliates from and against any claims, liabilities, damages, losses, costs, expenses, and reasonable attorneys' fees arising out of:
- the Customer's use of the Service;
- Submitted Materials;
- violation of these Terms;
- violation of applicable law;
- infringement of any third-party rights.
17. Suspension and Termination
The Company may suspend, restrict, or terminate any Account or Registration at any time if the Company reasonably believes that:
- these Terms have been violated;
- applicable law has been violated;
- continued use of the Service may expose the Company to legal or financial risk;
- fraudulent or abusive activity has occurred.
Termination of an Account does not affect rights and obligations that accrued prior to termination.
The Company may retain Registration Records where required by law, for dispute resolution, fraud prevention, security, or legitimate business purposes.
18. Governing Law and Jurisdiction
These Terms shall be governed by and construed in accordance with the laws of the State of Delaware, United States, without regard to its conflict of law principles.
Any legal action arising out of or relating to these Terms or the Service shall be brought exclusively in the state or federal courts located in the State of Delaware, and the Customer irrevocably submits to the jurisdiction of those courts.
19. Severability
If any provision of these Terms is held to be invalid, illegal, or unenforceable, the remaining provisions shall remain in full force and effect.
20. Entire Agreement
These Terms, together with the Privacy Policy, Refund Policy, and any other legal policies published by the Company, constitute the entire agreement between the Customer and SolCity Nav, LLC concerning the Service and supersede all prior understandings relating to the same subject matter.
21. Contact Information
SolCity Nav, LLC
651 North Broad Street
Middletown, Delaware 19709
United States
E-mail: scireg@scireg.org
Website: https://scireg.org
