The geology of western Peru. Cornish pioneers in South Africa. The defunct Eastern Counties Railway. Cheeses of the world. The jewellery of Hong Kong.
When Compass Books in Dartmouth, Devon, received an order earlier this summer for second-hand volumes on these subjects and many more, seller Lucy Hingston was delighted – if a little baffled.
And when she discovered that the buyer appeared to be a shipping company based at Heathrow airport, her suspicions increased.
Soon the 49-year-old was speaking to other second-hand dealers around the country – and discovered that Compass Books wasn’t alone.
There is now a growing realisation that Britain’s second-hand bookshops are being pillaged – not by voracious collectors of prized first editions but by Silicon Valley tech firms for a far more insidious purpose.
Today, the Daily Mail explains how millions of second-hand books are being bought in bulk and sent to warehouses across America, where the pages are sliced from the spine before being fed into industrial scanners, harvested for their text, and later destroyed.
This process, known as ‘destructive scanning’, is the latest method of gathering the vast reams of data being used by Big Tech to train its AI chatbots.
‘It’s a moral dilemma,’ says Lucy. ‘They’re my books, my babies. I feel like we’ve been eaten by the AI monster. It’s difficult.
‘I want the sale, but not to somebody who will b****r up everybody’s books.’
Pictured: Compass Books in Dartmouth, Devon, whose books have been pillaged by Silicon Valley tech firms to train AI
Pictured: A warehouse used by Anthropic to store books. Today, the Daily Mail explains how millions of second-hand books are being bought in bulk and sent to warehouses across America
Aside from the obvious concerns over copyright – of which more later – there are fears tech firms could be destroying rare books – even the last remaining editions of some historical tomes. As Lucy herself admitted, some of the books purchased from her store are ‘collectable and valuable’.
‘Bulk-buying books just to rip them apart to scan is shocking, but it’s just the tip of the iceberg when it comes to the brazen theft of copyright materials by the tech giants,’ says Dan Conway, CEO of the Publishers Association.
He adds: ‘They’re not just destroying books. Through their shameless actions – including the illegal use of content from books and journals without payment or permission – they are harming our world-leading creative industries and the livelihoods of authors, illustrators and journalists. This behaviour needs to stop.’
Last year, the Daily Mail launched a campaign, Make It Fair – backed by numerous celebrities and entrepreneurs including Sir Elton John, X Factor impresario Simon Cowell and Universal Music boss Sir Lucian Grainge – to stop Big Tech from stealing Britain’s creative genius.
This latest twist in the drive to satisfy AI’s insatiable hunger for content proves the imperative to resist is greater than ever. But with so many books already available online, why are tech firms turning to niche, second-hand bookstores and obscure volumes to train their AI models?
Court documents released earlier this year revealed that Anthropic – the AI giant behind the popular chatbot Claude – had been engaged in ‘destructive scanning’ since early 2024 as part of a secret initiative known as ‘Project Panama’.
‘We don’t want it to be known that we are working on this,’ declared one employee in an internal memo.
Court documents showed the firm ‘spent many millions of dollars to purchase millions of print books’ and then ‘stripped the books from their bindings, cut their pages to size, and scanned the books into digital form – discarding the paper originals’.
Join the discussion
Should tech giants be allowed to destroy old books just to feed their AI algorithms?
The books are bought by the AI companies and then destroyed after they are finished with them
Lucy Hingston (right) with comedian Susan Calman. When Ms Hingston received her order for a large volume she was delighted but became suspicious very quickly
AI ‘large language models’ – such as Claude and OpenAI’s ChatGPT – are ‘trained’ on vast amounts of text, audio, pictures and film footage. Type a ‘prompt’ into the chatbot, such as ‘tell me about the geology of western Peru’, and the response, and the rest of the conversation with the bot, will be based on the incomprehensibly huge bank of material it has previously absorbed.
This will include, for practical purposes, more or less everything available on the searchable internet, but also enormous amounts of pre-internet-era material that has already been digitised.
Now the companies are turning to books that have not previously been uploaded to the internet to widen their databases further.
Previously, Anthropic had trained Claude on more than seven million digitised books without paying a penny to the authors or publishers.
The company did so by sourcing the books from ‘pirate’ websites on which copyrighted material is illegally shared.
Five million books were taken from one of these sites alone.
Earlier this month a federal judge in San Francisco approved Anthropic’s $1.5billion (£1.1billion) out-of-court settlement following a class-action lawsuit filed by three affected authors in California in 2024. It was the documents from this case that revealed the firm had since pivoted to buying physical books.
Doing so avoids copyright infringement issues or illegality because under American law’s ‘first sale doctrine’, established by the Supreme Court as long ago as 1908, the buyer of a physical product can do whatever they want with it, including throwing it away.
Most book-lovers would find the practice of destroying huge numbers of physical volumes stomach-churning. Scenes of grinning young Nazis burning towering pyres of ‘un-German’ books in university cities across the Third Reich in 1933 are seared in the public memory
In other words, if Anthropic buys a physical book, it is legally – though not, many would argue, ethically – entitled to use it to ‘train’ Claude by scanning and destroying the physical copy.
This also means that any future user of Claude will be able to use the information contained in the now-destroyed book more or less for free.
Most book-lovers would find the practice of destroying huge numbers of physical volumes stomach-churning. Scenes of grinning young Nazis burning towering pyres of ‘un-German’ books in university cities across the Third Reich in 1933 are seared in the public memory: a ‘cultural genocide’ that continues to horrify us.
Of course, the AI companies would argue they are doing something different: rather than destroying books to censor the material they contain, as the Nazis did, they are destroying them in order to disseminate their contents more widely.
Nevertheless, the thought of those once-loved, out-of-print volumes being shredded for the financial benefit of faceless tech giants is deeply dystopian.
As the recent court documents explained, Anthropic has been seeking to train its AI models on ‘good’ writing in order to create better chatbots for its customers.
But rather than hire writers to create high-quality sample writing, the firm reasoned that it would be cheaper to purchase existing books.
Hence the Silicon Valley giants buying millions of books from around the world, including Britain, to train their chatbots without paying anything more than the low cost of the second-hand volumes. And Anthropic is not the only one at it. An investigation by 404Media found that an Amazon warehouse outside Las Vegas known as VGT3 was also being used to scan incalculable numbers of second-hand volumes.
‘I work at VGT3 here in Vegas, and all we do is scan books,’ revealed one anonymous Amazon whistleblower on an online forum.
AI ‘large language models’ – such as Claude and OpenAI’s ChatGPT – are ‘trained’ on vast amounts of text, audio, pictures and film footage
Second-hand books on sale in London. The Daily Mail has established that tech firms are especially targeting second-hand dealers, where the price per copy can be in pennies or a few pounds at most
‘Some are assigned to cut books, and others go to receive where they get books and scan the bar codes.’
Leaving aside the question of whether or not books should be scanned at all, the practice of cutting off the spines of books and destroying them afterwards has caused uproar.
So much so that even Elon Musk took to X last month to assure supporters: ‘I’ve asked the SpaceXAI team to preserve any rare books in a library and scan them the hard way [physically turning the pages] vs just cutting off the spine and scanning.’
It is, of course, possible to scan books in such a way that the text and binding are preserved – assuming the practice of feeding this information into the insatiable AI models is a worthy pursuit at all. But cutting the pages out first is ‘significantly faster’ and guarantees ‘shadow-free scans’, according to specialist scanner eRecordsUSA.
So which books do tech firms want most, and which bookshops are being targeted?
The Daily Mail has established that tech firms are especially targeting second-hand dealers, where the price per copy can be in pennies or a few pounds at most.
However, only shops with online catalogues appear to have been approached.
Intriguingly, the firms appear to be interested only in books published before 2022, the year in which generative AI achieved mainstream popularity following the launch of ChatGPT.
Books published before that period are highly likely to be free from AI-written text – whereas many authors of more recent books are, whether they admit it or not, using AI, while many books available for purchase online are entirely AI-written.
Training AI models on text written by AI can lead to what experts call ‘model collapse’.
Shelves of books in Camden Market. The thought of those once-loved, out-of-print volumes being shredded for the financial benefit of faceless tech giants is deeply dystopian
‘It’s the AI equivalent of a feedback loop gone wrong,’ explains Yarin Gal, Associate Professor of Computer Learning at Oxford University. ‘The more models feed on their own output, the further they drift from reality. Model collapse threatens to create an AI echo chamber.’
In other words, chatbots need to be trained on high-quality, human-written text. That means books published before the arrival of AI are guaranteed to be an effective ‘dataset’.
Which brings us back to the boom in second-hand book sales across Britain. Finn Windle, 30, works at The Minster Gate Bookshop in York. ‘We had an unusual batch order in May, quite a lot of books, and none of them were related subjects,’ he says.
‘It was quite confusing. There was local history, geology, a real mix of things, academic books and books for the general reader. Generally, non-fiction.
‘The purchaser was in the USA: Zoom Books. They’ve only ordered twice. Other people had had orders from them, so they seemed legitimate. But thankfully it’s gone quiet since.
‘If we knew destructive scanning is what they were doing, we wouldn’t have authorised the order. We’re not pleased about it – it’s awful.
‘Some of these titles aren’t even in print any more and there are no other copies available.’
Other booksellers as far afield as Germany, New Zealand and Australia have reported receiving orders from Zoom Books. The company has since denied being involved in destructive scanning. However, this does not rule out Zoom Books being the middleman responsible for collecting books for an AI giant.
It’s a similar story in Felixstowe, Suffolk, where Helen Bott, 64, runs Treasure Chest Books. ‘It started in early May,’ she says. ‘First, we got an order from Zoom Books in America. We got two or three orders for five or six books, and then a week later there were massive orders for 340 books, all to this place in America called Zoom Books. We’d never heard of it. I specialise in transport and railway books and the sheer randomness of the orders was noticeable. Fifty per cent were on railways, then Latin, or school history, or shipping companies.
‘We noticed these were all non-fiction. It felt like they were looking for unusual books: a history of a village in Devon, for example. You don’t want to see books destroyed. But I’m conflicted about it. From a business point of view, it’s good. But destroying books, that is the last thing we want to see.’
Meryl Halls, managing director of the Booksellers Association, agrees. ‘The potential scenario of tech companies bulk-ordering second-hand books to train their large language models is a concerning development,’ she says. ‘If true, it is another example of the lengths some AI companies may go to in order to access copyright-protected material without transparency or compensation.’
However, not everyone shares her concerns. ‘We have been joking and hoping that that would happen to us!’ says Patrick Kelly of Book Mongers in Brixton, south London. ‘We could use the business. My suspicion is that you need a catalogue of your inventory [for the AI companies to buy from you] – something we don’t have. To be honest, I don’t understand why the shops affected are complaining.’
And this attitude gets to the heart of the dilemma. For booksellers operating on tight margins, selling hundreds of books to AI firms makes sense. But, as every book-lover knows, Faustian pacts don’t end well.
A spokesperson for Anthropic told the Daily Mail: ‘Claude is trained on a mix of publicly available web data, commercially acquired datasets, and data we generate ourselves.
‘Sourcing books is a widely used approach for training large language models across the AI industry. None of our data acquisition programs buy and destroy rare or antiquarian books.’
A spokesperson for Amazon added: ‘Amazon purchases books through commercial channels to help develop and improve the products and services our customers use.’
Another infamous mass destructions of books was the ‘Bonfire of the Vanities’ in Renaissance Florence, when the friar Savonarola presided over the burning of books deemed immoral.
A new wholesale destruction of books appears to be underway again, some of which will never be printed again.
This time, however, the motivation is purely to enrich technology companies already enjoying wealth far beyond Savonarola’s dreams.
You must be logged in to post a comment Login