Anna's Archive claims Anthropic destroyed millions of scanned books
Anna's Archive, a self-described shadow library, published a guest post on its blog on 2026-08-05, written by a volunteer identified only by the pseudonym "u" and translated from Chinese. The post claims that several AI companies are secretly acquiring large quantities of secondhand books through intermediaries, then scanning and destroying them to obtain training data "untouched by machines" from before 2022, locking that material away on private corporate servers. Only one company is named: Anthropic.
According to the post, Anthropic launched a highly confidential effort called "Project Panama" in early 2024 and has spent tens of millions of dollars purchasing millions of paper books. The post says Anthropic scanned the books, used them to train its Claude models, then destroyed them all, and that the project came to light through a $1.5 billion copyright settlement. No link, case name or other source for that settlement, or for Project Panama itself, appears anywhere in the post.
The author calls the practice legal but, in their own words, "an extremely serious crime against humanity." Three reasons are given for destroying the books rather than keeping them: doing so stops competitors from scanning and training on the same copies, it avoids legal risk, and destroying books is cheaper than lossless scanning. The outcome, the post argues, is that the AI company becomes the only party in the world holding a digital copy, so knowledge once shared through physical books ends up monopolized on private servers, even as these companies promise their products will make human knowledge more accessible.
Anna's Archive describes itself as the world's largest shadow library and frames its response as building a "digital library of Alexandria" alongside other shadow libraries. It is asking volunteers worldwide to scan books, journal articles, newspapers, magazines and rare materials from libraries and archives and upload them, reasoning that 10 million volunteers scanning one book each would preserve 10 million items. Small-scale contributors get recognition and lifetime membership; for large-scale scanning, Anna's Archive says it can help cover the scanning costs and offer other rewards.
The post adds that AI-generated content has made up more than half of newly published internet content since the start of 2025, and asks how human civilization could be preserved once AI has absorbed the last sentence a human wrote on paper and everything left online is AI's own words. It appeals to readers who want the memory of human civilization kept from being monopolized, who want future generations to read humanity's work for free, and who do not want publishers profiting while authors get little. The post frames book scanning as a race against both publishers, which it says could "completely block knowledge," and AI companies destroying the world's remaining books and papers, closing by asking for volunteers to scan and upload, buy books to scan, or donate.
Key facts
- Anna's Archive published the post on 2026-08-05, written by a volunteer using only the pseudonym "u" and translated from Chinese, claiming several AI companies are secretly buying, scanning and destroying physical books for pre-2022 training data; only Anthropic is named.
- The post says Anthropic's "Project Panama," launched in early 2024, spent tens of millions of dollars on millions of paper books that were scanned to train Claude and then destroyed, and that the project came to light through a $1.5 billion copyright settlement with no link or case name given.
- Three reasons are offered for destroying scanned books: blocking competitors from training on the same copies, avoiding legal risk, and cutting costs versus lossless scanning.
- Anna's Archive, which calls itself the world's largest shadow library, is recruiting volunteers to scan books, journals, newspapers and rare materials, offering lifetime membership for small contributions and help covering costs for large-scale scanning.
- The post cites AI-generated content exceeding half of new internet content since the start of 2025 as evidence that human-authored material could vanish from the public record.
Why it matters
The post turns a specific, serious allegation, that a major AI lab destroyed millions of physical books after scanning them, into a broader argument about who controls knowledge once books are digitized. If a book is scanned and then destroyed, whoever did the scanning becomes the only party with a copy, closing off any way to check an AI company's training data, or its outputs, against the original text. That is the dynamic Anna's Archive is organizing volunteers against, framing rare-book scanning as a race it could lose permanently rather than routine archival work.
Who it affects
Anthropic is the only company named as allegedly running a book-destruction project, making this a reputational matter specifically for that company, even though the post frames the practice as common across "several AI companies." Authors and publishers sit on both sides of the post's argument: it accuses AI companies of profiting from books without keeping them accessible, while separately faulting publishers for "making a fortune while authors receive little." Libraries, archives and rare-book holders are the direct target of the volunteer call, since the post specifically asks for material from those collections. Anna's Archive is itself party to the outcome: it says it can help pay for scanning and offers rewards for large-scale contributions, so the volunteer effort it describes also grows its own collection.
How to use it
The post describes a direct way to get involved: scan books, journal articles, newspapers, magazines or rare materials, then upload them to Anna's Archive. Small-scale contributors get public recognition and lifetime membership to Anna's Archive, and for large-scale scanning projects, the post says Anna's Archive can help cover scanning costs and offer other rewards. It frames the scale of the ask in volunteer terms: 10 million volunteers scanning one book each would preserve 10 million items, on the post's own figures.
How solid is it
The post's central allegation, that Anthropic ran "Project Panama" and was exposed through a $1.5 billion copyright settlement, comes with no link, case name or other citation anywhere in the text. The only figure given for books affected is how many were purchased ("millions"); no count is given for how many were actually destroyed, and no date is given for when any destruction happened, only when the project reportedly began (early 2024) and when this post was published (2026-08-05). The post says "several AI companies" are involved but names only Anthropic. The author is identified solely by the pseudonym "u," with no real name, credentials or institutional affiliation given, and the post itself states it was translated from Chinese. None of this confirms or disproves the allegation; it means the claim currently rests on a single, anonymous, uncorroborated source.
Risks and caveats
Because Anthropic is named specifically while the broader "several AI companies" claim is not, this piece is easy to repeat as settled fact about one company when it is, on its own terms, a single unverified account. The post that makes the allegation is also the appeal asking readers to volunteer, buy books to scan, or donate to Anna's Archive, so the claim and the fundraising ask are not independent of each other. If the underlying practice is real, the post's own framing, that destroyed books cannot be rescanned, means any genuinely rare or one-of-a-kind volumes caught up in it would be lost as physical objects, not merely made scarce.
“It's legally permissible, but ethically, it's an extremely serious crime against humanity.”
— the post's author, Anna's Archive volunteer "u"