As John Milton wrote in his 1644 prose polemic Areopagitica, “books are not absolutely dead things.” While centuries have passed since Milton’s words, books have retained their role as signifiers of change, even in the face of a mind-smooshing attention economy; agents for cultural shift as much as they are emblems of it. Trump brandishes a personalised “God Bless the U.S.A.” Bible in his political speeches; India’s government bans literature on the Kashmir conflict; celebrity stylists use books to accessorise outfits. Whether we realise it or not, books are still emotive objects in our collective consciousness; we attach value and currency to them, which is exactly why AI company Anthropic’s new mode of book-scanning and digitisation, “Project Panama”, feels so dangerous.
Anthropic, currently the foremost name in AI tech, is feeding its flagship large language processing mode, Claude volumes of text discreetly acquired from second-hand and rare book sellers. The purchases are made by third-party buyers in bulk, so the intention and line of command are entirely untraceable. The purchased items include antiquarian, out-of-print titles that are scanned and digitised under the code-named Project Panama, although Anthropic has officially denied the use of rare copies. The project is led by Tom Turvey, who emphasised the secretive nature of their work in a court document. Turvey explicitly stated that Project Panama is an “effort to destructively scan all the books in the world” and that their team does not “want anyone to know,” outside of the company.
Considering the significance of print media in our lives, that recent news of the destructive scanning methods used by AI company Anthropic has taken the internet by storm. All over social media, including Substack and Reddit, users say Anthropic’s actions are “cartoonishly evil” and compare the act to the book burnings in Germany during WWII or the burning of the Library of Alexandria.
The act of digitising books is not out of the ordinary. Databases like the Internet Archive and Google Books have spent decades scanning texts for people to access across the internet. Discarding the physical copy after scanning is almost always optional but, for Anthropic, the indifference towards the book is entirely intentional, even systemic. To destructively scan a book, a hydraulic press is used to cut off the spine and the edges for faster, page-by-page scanning. After that, the contents are shredded. In non-destructive processes, the spine is reattached after scanning, and the pages are put back as-is. For Anthropic, an AI company valued at $965 billion, this was a task that was allegedly too time-consuming and cost-intensive.
Before making this decision, Anthropic used pirated sources to train its AI model, which ended in a copyright class-action lawsuit and a $1.5 billion settlement. The evidence from the Bartz vs. Anthropic court ruling is what unveiled Project Panama to the public. The settlement was approved by the presiding judge, and the destructive scanning initiative was allowed as an alternative to using pirated sources, under the ‘fair use’ or ‘transformative use’ clause within US copyright law. This means that Anthropic is only allowed to make digital copies of the books they acquire as long as the physical text is destroyed – i.e the physical text is ‘transformed’ into digital scans.
This information pushes an uncomfortable question into view: doesn’t this clause make it that much more convenient for AI firms to hoard knowledge and commodify it? Take, for evidence, Sam Altman’s heady claim at the BlackRock summit in Washington DC, where he boasted his vision of “a future where intelligence is a utility, like electricity” and “people buy it [from AI firms] on a meter and use it for whatever they want to use it for”.
Considering the accumulating evidence, there has been significant pushback against Anthropic as a whole, and their adjacent “AI is the future” faction online. Social media users remark at the irony of AI tech being served up as the replacement for all cognitively laborious tasks while LLMs digest centuries of human writing.
It is no surprise that the internet has also armed itself with literature to explain how the world is falling right into dystopia. Mentions are being made of Ray Bradbury’s Fahrenheit 451 and George Orwell’s 1984, both of which are works of fiction that stay relevant in their warning against a coercive and militarised surveillance state where deviant ideas – often represented by books – are extremely dangerous and illegal. In both Bradbury’s and Orwell’s worlds, books are condemned to fire. In our reality, however, there is no incinerator and no smoke to point towards. Even book burnings, in today’s age, have been optimised into their quietest forms. What Anthropic does to millions of books, it does behind closed doors, under code-names and guises of goodwill.
Internet discourse is hinting towards a rising public consciousness as people try to uphold the sanctity of books. In some posts, users upload scenes from local libraries, likening the spaces to “real data centres” that need to be protected. Anthropic’s actions reveal, in a rather contradictory fashion, that human intelligence and human emotion are still the most sought-after resources, bar none, even centuries after Milton’s words. An AI model may regurgitate pieces of literature to create a semblance of something complete and artistic, but its output will only ever be a mimicry of what human hands have already built. In Anthropic’s vision, the written word, the physical medium which encapsulates the human condition, is guarded; It is encrypted, away from reach, into a large language model. Then, a second question arises: If we make the shift into a world that is purely digital, where literature is mediated by more sinister third parties than publishers, who will protect our Alexandria?
