Anna's Archive Calls for Saving Rare Books Before They're Destroyed to Train AI

Anna's Archive, which calls itself the largest open library in history, has published a guest post by a volunteer writing under the pseudonym “u” (translated from Chinese) that carries an alarming appeal. According to the authors, some AI companies are using intermediaries to buy up used books, scan them and then destroy them — all for training data “untouched by machines,” meaning data created before 2022.

“Project Panama”

As an example, they cite Anthropic's “Project Panama,” which surfaced, according to the author, in connection with the settlement of a $1.5 billion copyright infringement lawsuit. It's claimed that in early 2024 the company launched a closed project, spent tens of millions of dollars on millions of paper books, scanned them to train the Claude model, and then destroyed them. The author's outrage is that while this is technically legal, in essence, as they see it, it is a grave crime against humanity.

Why destroy the books? The author lists pragmatic reasons: to keep competitors from scanning the same texts, to reduce legal risks, and to save money — destroying a book is cheaper than carefully digitizing it. As a result, the company is left with the only digital copy in the world, and knowledge ends up locked away on private servers.

In this the author sees a paradox: while promising to “make knowledge accessible,” AI companies are destroying the most reliable carriers of that knowledge. The public gets smarter assistants at the cost of an enormous body of books vanishing from the public domain.

A call to volunteers

Anna's Archive's answer is to build a “digital Library of Alexandria” through the efforts of shadow libraries and volunteers around the world. The project asks people to scan and upload books, magazines, newspapers, and old and rare editions from any libraries and archives, especially the kind of material that is easily lost. The logic is simple: if everyone digitizes one book, and ten million volunteers sign up, that adds up to ten million saved editions. For small contributions they promise lifetime membership; for large-scale digitization, help covering scanning costs and rewards.

Separately, the author raises the alarm about the trend: since the start of 2025, they say, AI has generated more than half of all new content on the internet. If this keeps up, it will become impossible to tell what was written by a human, and once AI “swallows the last phrase written by people on paper,” only its own words will remain online.

The ending is written in a manifesto-like tone: this is a race against time, and the goal is to manage to scan all the world's publications before publishers close off access to knowledge for good and AI companies buy up and destroy the last books. The claims made in the text cannot be verified against independent sources — this is Anna's Archive's own position.

Source: Hacker News