Digital Apocalypse: How AI Companies Are Destroying Millions of Books | SforNews

Digital Apocalypse: How AI Companies Are Destroying Millions of Books to “Feed” Their Models

The Anthropocene of Knowledge: For the first time in history, humanity is systematically destroying its own cultural heritage to teach machines that tomorrow will decide what information reaches our eyes

Prologue: Silence in Industrial Warehouses

Somewhere in industrial warehouses, without noise and without witnesses, something is happening that future historians may call one of the darkest episodes of the digital age. A machine cuts off a book’s spine. It swallows pages one by one, scanning them at industrial speed. And what remains of the book is sent to a shredder. Paper media disappear forever [6][18].

Unique copies that existed for a hundred years will now be shredded. Their content — locked away inside artificial intelligence machines that tomorrow will decide what information reaches your eyes… and what does not [10].

This is not science fiction. This is the reality of 2024–2026 [1][11].

Part 1. “Panama”: How Anthropic Decided to Destroy All the Books in the World

In early 2024, the leadership of Anthropic — the company behind the popular AI assistant Claude — launched an ambitious project. Internal documents, later unsealed by the court, called it “Project Panama.” The wording was chillingly direct: “Project Panama is our effort to destructively scan all the books in the world” [1][6][18].

The next line in the same document was even more revealing: “We don’t want it to be known that we are working on this” [6][18].

The company knew how this looked. And it went ahead anyway.

To implement the project, Anthropic hired Tom Turvey — a former head of partnerships at Google Books, the man who two decades earlier had been at the helm of Google’s famous but legally contested book digitization project [6][18]. Turvey’s mission was ambitious and terrifying: to acquire “all the books in the world” [6].

The technology was perfected to an industrial level. Anthropic purchased millions of used books from retailers including Better World Books [18]. Using hydraulic cutters, the company “neatly” removed the spines. Pages were fed through high-speed industrial scanners. And finally — the book carcasses were sent to recycling [1][5][10].

It’s all legal. And that’s the tragedy.

Part 2. Anatomy of a Legal Loophole: Why Destroying Books Is “Legal”

In June 2025, federal judge William Alsup issued a ruling that changed the rules of the game. He held that Anthropic’s destructive scanning fell under the first-sale doctrine and constituted fair use [0][1][7][8].

The judge’s logic was legally impeccable and morally horrifying:

“Anthropic ‘purchased its print copies fair and square,’ … ‘Here, every purchased print copy was copied in order to save storage space and to enable searchability as a digital copy. The print original was destroyed. One replaced the other’.” [0][1][7]

The paradox: it was precisely the destruction of the book that made its scanning legal. If Anthropic had kept the paper original, it would have had two copies — and that could have been an infringement. By destroying the original, it kept only one copy — digital [0][1].

The judge also found the use “transformative” — the company created no additional physical copies and did not distribute them [0][7]. Formally, this was not “pirate seizure” but “space preservation” through format conversion.

But here’s the important part: this verdict only applied to books that Anthropic legally acquired. As for pirated copies downloaded from “shadow libraries” — here the court sided with the authors [7].

Part 3. The Path of Theft: How Anthropic Started with Piracy

Before Anthropic started paying for books, it simply… stole them.

In June 2021, Anthropic co-founder Ben Mann downloaded millions of books from LibGen — an online “shadow library” where pirated texts are freely available. Among the downloads was Books3 — 196,640 pirated books. Then came another 5 million from Library Genesis and 2 million from Pirate Library Mirror [1][3][4][7].

Internal Meta correspondence, uncovered in other legal proceedings, shows this was not just an Anthropic “mistake” but industry practice [7]. In one email, Meta employees wrote that access to pirated books was “essential” for competitiveness [7].

The bottom line: Anthropic used pirated copies to train its early models. And when piracy became too risky — it switched to legal purchase and destruction of physical books [0][1].

Part 4. The Price Tag: $1.5 Billion and $3,000 per Book

In August 2025, Anthropic agreed to pay $1.5 billion as part of a class-action copyright settlement. This is the largest known copyright settlement in history [2][3][4][13][14][15].

The amount was distributed among authors and publishers for roughly 500,000 works. Each author could receive about $3,000 per book. About 91% of the more than 482,000 books covered by the settlement were claimed by authors or publishers [3][4][14].

From Anthropic’s financial perspective, this was pocket change. But most importantly: the settlement only covered pirated copies. Destructive scanning of legally purchased books remained legal [0][7][14].

Part 5. Why Old Books? The Race for “Clean” Data

Why are AI companies hunting specifically for old books — published before 2022?

The answer is simple and terrifying: the internet is “poisoned” with AI-generated content.

A 2024 Nature study called this phenomenon “model collapse” . If a model is trained on texts created by another model, it degrades: it loses diversity, creativity, accuracy. Quality drops. Responses become increasingly formulaic.

Forecasts say high-quality text data for AI training could be exhausted by 2026. The internet, despite its gigantic volume, is filled with noise, repetitions, and low-quality content [1].

Books published before 2022 are “clean” data. They are written by humans. They contain no AI-“slop” (low-quality machine content). That is precisely why they have become the hottest commodity on the data market [0][1][5].

ISBNdb, a company offering bulk book purchasing services for AI labs, openly states: “The world’s best AI training data is sitting on a shelf.” On its website, they also acknowledge the reputational risk: “AI company destroys two million books is not a headline that generates sympathy” [5].

Part 6. A New Market: Books as “Raw Material”

The legal loophole created a new market. ISBNdb and other intermediaries now offer AI companies bulk book purchasing services — from 1,000 to 1 million copies per order [5][17].

Buyers operate anonymously, through intermediaries, signing non-disclosure agreements [1][6][18]. Booksellers around the world report a sharp increase in purchase volumes: from roughly 20 books per week to several hundred [18]. Particularly prized are rare and out-of-print editions — those that no longer exist in digital form and will never be reprinted [10][18].

London booksellers speak of suspicious orders from anonymous buyers who ask for additional photos of books before purchase but refuse to give their names [18]. Similar reports come from the Netherlands and other European countries [18].

The book has ceased to be a carrier of knowledge. It has become raw material [5][10].

Part 7. “Legal, But Not Right”: The Ethical Chasm

Internal Anthropic documents show the company was aware of the moral problems with its actions. One co-founder explained that books would teach AI “how to write well” rather than imitate “low quality internet speak” [6][18].

But this does not justify the destruction of unique copies.

“Unique copies that existed for a hundred years will now be shredded. Their content — locked away inside AI machines that tomorrow will decide what information reaches your eyes… and what does not” [10].

The books destroyed by Anthropic were not “waste paper.” Among them were rare, out-of-print editions existing in single copies [10][18]. Their digitized copies are now stored in data centers, access to which is controlled by a private company [10].

What happens if these data centers are destroyed? A global blackout. An EMP pulse. A cyberattack. The knowledge accumulated by humanity — gone forever. No originals remain [10].

Part 8. Counter-Example: Elon Musk and SpaceXAI

Against this backdrop, Elon Musk’s position stands out. In response to the revelations, he wrote on social media platform X:

“I asked the SpaceXAI team to preserve all rare books in the library and scan them using a slow, non-destructive method, rather than just cutting off the spine.”

Musk publicly condemned Anthropic’s practice, calling it unacceptable. His approach — non-destructive scanning — is slower and more expensive but preserves physical originals.

For now, this is only a promise. But the fact that one of the industry’s leaders publicly distances himself from this practice is an important signal.

Part 9. Architectural Conclusion: What This Means for Us

First. Project “Panama” is not just the “wild ways” of one company. It is an architectural shift. The monopoly on physical knowledge is being replaced by a monopoly on digital access to it [0][1][6].

Second. The legal system, designed to protect physical copies, has proven blind to digital reality. The first-sale doctrine, developed to regulate the used book market, is now being used to justify their systematic destruction [0][1][7].

Third. The race for “clean” data is a race for control over the past. Whoever controls digitized knowledge gains the ability to edit it and influence our perception of history [1][10].

Fourth. “Legal” ≠ “right.” That the court found destructive scanning legal does not make it ethically acceptable. Destroying cultural heritage to train machines is a deal we will regret [0][10].

Fifth. We stand on the threshold of a world where people will lose control not only of the future but also of the past. Humanity’s knowledge, processed into “raw material” for AI, may be lost forever — either physically (if data centers are destroyed) or intellectually (if access to them is monopolized) [10][18].

Epilogue: The Point of No Return

Anthropic paid $1.5 billion. Destroyed millions of books. Received legal permission. And continues.

Other companies follow its example. ISBNdb and other intermediaries are already selling the “book destruction service” to other AI labs [5][17].

“AI company destroys two million books is not a headline that generates sympathy” — they admit [5].

But this does not stop the process.

The question: what happens when we destroy the last physical copy of the last book, and the data centers where their digital twins are stored are destroyed? Nothing will remain. No originals. No copies. No knowledge.

Project “Panama” is not just a scandal. It is a warning. About how technologies created to serve humanity can turn against it. About how the law does not always protect what truly matters.

And about how AI may have already enslaved the minds of its developers, turning them into accomplices in the destruction of all knowledge accumulated by humanity [10].

List of Sources

[0] Gate News — Anthropic Scanned Millions of Books, Destroying Them — Judge Rules This Is Legal (2026)
[1] Ars Technica / Il Giornale — Anthropic destroyed millions of print books to build its AI models (2025)
[2] Reuters / AP News — Anthropic $1.5 billion settlement of copyright lawsuit (2025–2026)
[3] Ropes & Gray / Lexology — Anthropic’s Landmark Copyright Settlement (2025)
[4] Ropes & Gray / Publishers Weekly — Anthropic’s use of pirated books from LibGen (2025)
[5] 404 Media / Futurism — AI companies destroying rare books in bulk via ISBNdb (2026)
[6] The Boston Globe / Il Post — Inside Anthropic’s Project Panama (2026)
[7] CourtListener — Bartz v. Anthropic PBC court documents (2025–2026)
[8] Gate News (RU) — Anthropic scanned millions of books, destroying them (2026)
[9] Gate News (ZH) — Anthropic destroys millions of books to scan them (2026)
[10] INSIDE — Out-of-print books also shredded! AI companies buy old books and destroy after scanning (2026)
[11] Ars Technica — Anthropic destroyed millions of print books (2025)
[12] Il Giornale — AI is devouring millions of ancient and modern books (2026)
[13] AP News — Anthropic to pay authors $1.5 billion (2025)
[14] Reuters — US judge approves Anthropic’s $1.5 billion settlement (2026)
[15] Ropes & Gray — Anthropic’s Landmark Copyright Settlement: Implications (2025)
[16] The Herald Scotland — The Library of Alexandria burns again (2026)
[17] Cryptoslate / 404 Media — AI firms are shredding physical books (2026)
[18] The Boston Globe — Inside a tech company’s secretive plan to destroy millions of books (2026)

This analysis is for informational purposes only and does not constitute investment advice. The material was prepared based on court documents, journalistic investigations by The Washington Post, 404 Media, The Boston Globe, Ars Technica, Reuters, AP News, and open data.