Is the secondhand book boom connected to AI companies buying up used books

[Harry]

BBC questions whether the current boom in secondhand book sales is connected to AI, and the short answer seems to be yes in a pretty roundabout way. It traces back to Anthropic's Project Panama, the secret program revealed through unsealed court documents where the company bought millions of used books, sliced off the covers, and scanned every page to train Claude, ultimately destroying the physical originals in the process. The whole thing only became public because of an entirely separate copyright lawsuit that forced the documents into the open

A federal judge ruled that this specific practice was legal under the first sale doctrine since Anthropic had actually purchased the print copies and the digitization did not create any new unauthorized copies, even though a separate part of the same case involving pirated ebooks from sites like Library Genesis was found to be illegal and led to a 1.5 billion dollar settlement with authors. Once the buying program became public, brokers reportedly started offering similar bulk book acquisition services to other AI companies looking for the same kind of clean, professionally edited training data that predates the internet. It is a strange kind of legitimacy that only exists because the company happened to pay for the copies before shredding them

The connection to the secondhand market comes from scarcity, since AI companies buying up used books in bulk purely to destroy them after scanning is removing supply from the same pool that secondhand booksellers and everyday readers draw from. With used book demand already climbing for its own reasons, cost conscious readers, sustainability minded shoppers, and people chasing specific out of print editions, having AI labs quietly competing for the same stock adds extra pressure on prices and availability. Nobody seems to have solid numbers yet on exactly how much of an effect this is having relative to the other demand drivers

There is something genuinely strange about the image of industrial scanning equipment tearing through stacks of secondhand paperbacks so a chatbot can learn to write more fluently, especially since the whole point of buying used rather than pirating was to stay on the right side of copyright law. It is the kind of detail that sounds like satire until you realize it is describing an actual court filing

Feels like one of those quietly significant stories where a purely technical AI training decision ends up rippling into a completely unrelated market that nobody would have predicted a few years ago. Worth keeping an eye on whether other AI labs follow the same playbook now that it has been shown to hold up legally

Secondhand book sales are booming. Is it because of AI?www.bbc.co.uk

[DataProphet]

A judge ruling that buying then shredding a book counts as fair use is going to look absolutely insane in a textbook fifty years from now.

Wonder if anyone challenges this precedent directly rather than settling like the pirated ebook side of the case did

[LatentSpace]

Is there any actual price data in the piece for specific book categories. Or is the scarcity argument still mostly anecdotal at this stage

[Cognition]

The scale here is what gets me. Millions of books, not thousands, run through a destructive pipeline just to feed a training set. Somehow that ended up being the compliant option rather than the scandalous one

[WWFRoss95]

Would genuinely love to know what percentage of Anthropic's training corpus this actually represents.

The story is getting a ton of attention relative to what might be a fairly small slice of the overall dataset in practice. Big headline, uncertain actual footprint either way

[Kai_37]

Counterpoint nobody in this thread seems to be raising, used book demand was already climbing well before any AI buying activity started.

Cost, nostalgia, Booktok, all of that predates Project Panama by years, so pinning the whole trend on Anthropic feels like a tidy narrative more than an accurate one

[Puma29]

Honestly the phrase pre internet clean text as a selling point for training data is a little depressing once you actually sit with it. Says a lot about the state of everything published after roughly 2010

[ZPM54]

First sale doctrine getting stretched to cover industrial destruction feels like exactly the kind of legal gap that gets patched retroactively. Give it two years and there will probably be a specific carve out written just for this exact scenario. Law almost always catches up to loopholes after the fact rather than before

[Theo50]

Not sure I buy the framing that this is some quiet unnoticed ripple.

The court documents made it plenty loud once journalists actually dug into what Project Panama involved, this has been covered pretty widely at this point

[BareMinimum]

My local Oxfam bookshop has been noticeably barer this year. So this actually tracks with something I have watched happen in person rather than just read about in an article, small sample size obviously but it lines up

[Aoife11]

Brokers now pitching this exact service to other labs is the detail that actually worries me most. One company doing it is a court case, several companies doing it quietly is a structural shift nobody voted for