Artificial intelligence companies are reportedly acquiring millions of used books from second-hand dealers, scanning their contents to train AI models, and subsequently destroying the physical copies. This practice, described as "destructive scanning," has raised ethical concerns among booksellers and publishers who fear the potential loss of rare and valuable texts.
AI models, such as Anthropic's Claude and OpenAI's ChatGPT, require enormous amounts of text data to learn. While much of this data comes from the internet, companies are increasingly turning to digitizing pre-internet era books to broaden their datasets. This approach bypasses copyright infringement issues in the United States due to the "first sale doctrine," which allows the buyer of a physical product to do as they please with it, including discarding it.
Court documents revealed that Anthropic has been engaged in "Project Panama" since early 2024, involving the purchase of millions of print books. These books are then stripped of their spines, scanned, and the original copies are discarded. Previously, Anthropic had trained its AI on over seven million digitized books sourced from pirate websites, a practice that led to a $1.5 billion out-of-court settlement following a class-action lawsuit by authors.
This new method of acquiring physical books allows AI companies to obtain content without direct copyright concerns, though many argue it is ethically questionable. Booksellers have reported receiving unusually large and diverse orders from entities like "Zoom Books," a company that has denied involvement in destructive scanning but may be acting as a intermediary.
Publishers are also voicing concerns. Dan Conway, CEO of the Publishers Association, stated that "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." He added that such actions harm creative industries and the livelihoods of authors.
An investigation by 404Media also indicated that an Amazon warehouse near Las Vegas, known as VGT3, is being used for scanning second-hand books. An anonymous Amazon whistleblower confirmed that the facility's primary function is scanning books, with some employees assigned to cut them and others to scan barcodes.
Tech firms appear to be targeting books published before 2022, the year generative AI gained mainstream popularity. This is to ensure the training data consists of human-written text, avoiding "model collapse," a phenomenon where AI models trained on their own output drift from reality. Books predating widespread AI use are seen as a reliable source of high-quality, human-generated content.
While some booksellers, operating on thin margins, welcome the business, others express distress. Lucy Hingston, a bookseller from Dartmouth, described the situation as "difficult" and felt like "we've been eaten by the AI monster." She questioned the ethics of destroying books, even for profit.
Despite the controversy, spokespersons for Anthropic and Amazon defended their practices. An Anthropic representative stated that sourcing books is a common industry practice and that they do not buy or destroy rare or antiquarian books. Amazon indicated that book purchases are made through commercial channels to improve their products and services.
The practice has drawn parallels to historical book burnings, such as the "Bonfire of the Vanities" in Renaissance Florence, although the motivations differ. While past instances involved censorship or moralistic agendas, the current situation is driven by the commercial imperative of technology companies seeking to enrich their AI models.