AI, Book Burning & The Mandela Effect
On the ownership and editing of knowledge.
Protein, 17th September 2026
What began as rumours in 2024 came to light this summer: AI companies are buying up rare books, scanning and then destroying them; it being cheaper and quicker to slice the spines off books than preserve them. A recent investigation by 404 Media, in which they attached an AirTag to an order from a rare book seller, revealed the package ended up at an Amazon AI training facility, which houses VGT3, an operation which represents itself with an image of a dinosaur reading, or ripping apart, a book. Anonymous booksellers told 404 Media these bulk orders of rare texts had been accumulating in a very orderly fashion: it seemed buyers were methodically working through the serial numbers uniquely assigned to published texts.
Amazon has been digitising books for decades, but this practice points to something new, something more insidious. Already, AI limits the information it provides us with, which is highly susceptible to reinforcing the biases entrenched in culture and society. Slowly, and then seemingly all at once, these systems are dictating what knowledge is available to people; something which these recent revelations only further support.
The acquisition and destruction of books picked up pace after AI companies came under fire for copyright infringement. Anthropic’s Project Panama, a previously secret mission, was exposed in Bartz v. Anthropic: a $1.5 billion class action lawsuit on Anthropic’s unauthorised use of copyrighted books to train its AI models. As court exhibit 21, an internal memo by Anthropic, reveals: ‘Project Panama is our effort to destructively scan all the books in the world’. Encouraging total discretion, the memo continues, ‘Why use a codename? … [B]ecause we don’t want it to be known that we are working on this.’ The question is why exactly.
One potential answer comes from SEED CLUBBER Julián Medrano Hoyos: ‘Books are one of the most powerful objects humans have ever invented and this to me is a clear example of how these big tech firms are trying to monopolise human knowledge.’ In response, Amir Gamble suggests combatting this has to come from collective resistance: ‘I think it's a public duty to push back on stuff like this. Not just with activism, but learning to be archivists collectively.’ Physical (or server-based) media as both rebellion and archivism is precisely what Willem Deisinger examines in the SEED Piracy’s Back. Pirated material was almost entirely streamed in 2023, but by 2024, 50 per cent was downloaded, demonstrating a desire for media ownership, even if acquired illegally.
AI companies' procedural demolition of books is well underway, and unlikely to stop anytime soon. Considering this, Gamble suggests: ‘I think in the tradition of real open source and tech, they should make all their scans public, or create some kind of open library if this continues’. Indeed, if all books, and therefore knowledge, is digitised and physical copies destroyed, this knowledge becomes vulnerable to monopolisation and manipulation. Nothing is printed in black and white, and everything can be edited, updated or erased at any time. It would be naive not to think of Project Panama and similar schemes as blatant (albeit initially secret) iterations of the book burning and censorship of Nazi Germany or Red Scare. Even today, restricted access to information remains a critical issue, which the recent protests in Nepal shone light on.
‘It’s interesting thinking about the monopolisation of knowledge. I sometimes get to quite an extreme place with this regarding control’, Sophia Epstein says in the SEED CLUB discussion. ‘I’m not suggesting AI companies are going to be editing books like Stalin used to edit photos, but there’s just something weird to me about there only being digital (and editable) versions of books’. But books are already being edited, however, Epstein highlights, noting the bizarre references to TikTok Pretty Little Liars books now include, despite being first published in 2006, a whole decade before TikTok was founded.
As Angelina Mazza explains in The New York Times: ‘In publishing, the practice of updating cultural and technological references in older books is called modernization.’ This is something that mostly occurs in Young Adult fiction, and is separate from sensitivity editing, which changes language considered offensive, such as the revisions of Roald Dahl’s books. While the modernisation of texts is more about relevancy than eradication, in fairly innocent terms, it does illustrate how vulnerable books are to alteration.
Another example Epstein shares is Ryan Gosling’s flicked-out wrist in the La La Land poster. To mark the 10th anniversary of the film, Gosling’s hand was ‘corrected’ and updated on the poster. ‘This sucks. The gay little wrist ruled.’ Jason P. Frank writes in Vulture. One Redditor comments: ‘This is gonna be evidence of the Mandela Effect for so many people in five years’.
In a similar vein, as writer Kristin Robinson outlines in an Instagram video, The 1975 decided to remove the track ‘Human Too’ from the band’s album Being Funny in a Foreign Language. ‘An album in the age of streaming is a very loose term’, Robinson explains. Updating or removing music in this way can feel like historical negationism: the denial or distortion of history.
Considering that idea of the Mandela Effect in more depth, the control of books and media by AI companies or streaming platforms can cause us to doubt our own memories or produce false ones. AI hallucinations – chatbots offering us false information as fact – are already a well-known risk, but becoming harder to isolate when everything is questionable. The ownership and archiving of physical media is about nostalgia and resistance, but now also related to what and how we remember.
Comments