A complaint filed with the Federal Trade Commission this week asks a question that copyright litigation has never quite reached: what happens to competition when the raw material of artificial intelligence is destroyed after use. Civil-society groups urged the agency to investigate AI companies that buy physical books, scan them, and discard the paper — arguing that destructive scanning eliminates scarce titles from the used-book market and converts a shared cultural commons into a private corpus no rival can obtain. [1]
The claim's physical evidence is anecdotal but accumulating. Scott Brown, an antiquarian bookseller writing Friday, estimates tens of millions of books have already been consumed this way, describing bulk buyers stripping shelves of obscure titles that exist in few other copies. [2] Anna's Archive, the shadow library, has called for volunteers to scan rare books before they vanish into servers its organizers describe as permanently private. [3] The practice has a legal pedigree: federal courts have held that scanning lawfully purchased books for internal training use can qualify as fair use. Destruction was never the holding. It is the side effect nobody litigated.
Hannah Arendt observed that destruction is easier than construction, which is why it attracts less scrutiny. The FTC filing inverts her observation usefully. If a company scans a book and keeps it, authors have a live grievance and lawyers have a forum. But if a company scans a book and pulps it, two harms compound silently: the reader's copy is gone forever, and every future competitor must train on something lesser. Fair-use doctrine asks whether the copying harms the market for the original. It has no category for eliminating the original.
The epistemics here deserve as much attention as the economics, because the two sides are mispricing certainty itself. On X, library and archive communities treat the destruction as established fact — the viral formulations write themselves, a library burned to own the tokens. Mainstream coverage hedges the scale claims into obscurity; Axios reports the advocacy without endorsing the arithmetic, and no outlet has independently audited how many titles have actually been lost. [1] Both registers miss the point that survives either number. Even one destroyed unique title is a permanent subtraction from the public record made for a private gain — the injury does not require millions to be legible, only irreplaceable.
The booksellers' testimony carries the weight precisely because it is not ideological. Dealers who have spent decades tracking obscure imprints noticed algorithmic buying patterns before any advocacy group organized: single orders sweeping up decades-old guides, manuals, local histories — the long tail that Amazon's listings barely index and no digital archive reliably preserves. When the paper copy goes, existence itself becomes unverifiable. A book that survives only inside a corporate training corpus is, for every purpose except the corporation's, a book that has ceased to exist. [2][3]
Whether the FTC takes the case will say something about whether American competition doctrine can see physical infrastructure anymore. Antitrust learned to count data as an asset; it has not yet learned to count the destruction of data's substrate as a barrier to entry. The groups pressing the commission are asking it to notice that training-data moats were built partly out of shredded paper.
The libraries understood something the platforms did not: preservation is a practice, not a byproduct. A civilization that outsources its memory to entities with no preservation duty should not be surprised when the memory arrives with terms of service attached. [3]
-- ANNA WEBER, Berlin