The Books Being Destroyed Are Not Rare. They Are Out of Print.

The rarity claim entered through a 404 Media report on 21 July 2026, which described a surge of anonymous bulk orders reaching secondhand booksellers. One seller said his weekly volume went from about twenty books to several hundred. He said his inventory included uncommon and out-of-print stock, and he worried about where it was going.

Read carefully, that is a bookseller’s inference, and the reporting labelled it as one. He did not know who was buying. Nobody in the chain claimed to have identified a last surviving copy. By the time the story reached a viral thread on 27 July and the aggregators that followed it, uncommon and out of print had hardened into rare, then into antique, then into titles with no other known copies. That escalation happened in the retelling, not in the record.

There is a technical point that resolves this cleanly. The bulk orders described by booksellers are keyed to ISBNs. The bookseller who spoke to 404 Media said so explicitly, noting that every title in the surge carried one. The buyers are not walking shelves and selecting by scarcity. They are sending spreadsheets of identifiers.

The International Standard Book Number was standardised as ISO 2108 in 1970, following a British numbering scheme introduced in 1967. Books published before 1970 do not have ISBNs and never will. Estimates of the pre-1970 corpus run to tens of millions of editions. The entire body of printed human culture up to the moon landing sits outside the ISBN system entirely. So an ISBN-keyed purchase order cannot, by construction, reach genuinely antiquarian material. Whatever is in those spreadsheets, it is not an incunable, not a nineteenth-century first edition, and not a pre-war monograph. The selection mechanism forecloses it.

Until this week the argument stayed abstract, because no one outside the buyers had seen an actual list. On 31 July 2026, Fortune published one. Pieter de Vries, an antiquarian bookseller in Haarlem, received an email from a company identifying itself as 2077AI, requesting quotes on a spreadsheet of 3,001 ISBNs with shipping to China. He assumed it was phishing and ignored it. When a journalist later showed him what it was connected to, he provided the file.

The titles were mostly published in 2020 and 2021, by Emerald, Elsevier, Wiley, Routledge and Oxford University Press. The subject range ran across business, education, engineering, public policy and medicine. This is recent academic and professional publishing. It is the least rare category of printed matter in existence. The single documented procurement request in the public record is a list of five-year-old textbooks.

Here is the part that matters for anyone working in governance, because the instinct to defend the dramatic version is exactly backwards. If your thesis is that rare books are being erased, you lose on the facts, and the real problem dies with the thesis. If your thesis is that millions of legally purchased ordinary books were converted into a private, unauditable corpus, you are on solid ground and the problem gets larger rather than smaller.

Consider the contrast with Google Books. That programme scanned largely non-destructively and produced a public-facing index, and it survived its own copyright challenges partly on the grounds that its output was accessible and searchable by the public. Anthropic’s programme is the inverse: the books are gone, the corpus is private, and the only external visibility into what was ingested comes through litigation. That is a legitimate governance concern. It does not require the books to have been rare.

The $1.5 billion settlement addresses the piracy claims. It does not address the question of what a private company may do with a corpus assembled from the destruction of lawfully purchased physical copies of the world’s recent academic and professional literature. That question is open, and it is worth asking on accurate terms.