The short version
- A recent US court decision allowed AI firms to use purchased books for training without violating copyright, sparking a wave of bulk acquisitions from independent sellers.
- Internal documents reveal Anthropic’s 'Project Panama' aims to destructively scan global book stocks, a process that involves removing spines and recycling the physical remains after digitization.
- Booksellers express mixed feelings about the financial boost but worry about the permanent loss of rare editions and the ethical implications of destroying cultural artifacts for machine learning.
Independent booksellers across multiple countries are reporting a sudden and significant increase in bulk purchases of secondhand novels, creating a puzzling trend within the used book trade. Stuart Manley, who operates Barter Books in Northumberland, noted that a single order from a Canadian company recently matched his typical weekly sales volume of two to three thousand titles. This level of demand is unprecedented for retailers with decades of experience, as such massive orders lack the usual patterns associated with individual readers or library acquisitions.
The prevailing theory among industry observers is that these purchases are not driven by human readers but by artificial intelligence companies seeking data. The surge appears linked to a 2025 US court ruling in which Judge William Alsup determined that Anthropic’s use of authors’ books for training its AI software did not violate copyright law. The judge characterized the usage as 'exceedingly transformative,' thereby permitting it under existing legal frameworks. This decision effectively opened the door for AI firms to acquire physical books legally and at scale.
Court documents unsealed last month provided further insight into how these acquisitions are utilized, revealing that Anthropic engaged in a practice known as destructive scanning. Internal communications referred to this initiative as 'Project Panama,' with stated goals of ingesting all books globally. Destructive scanning involves shipping physical volumes to industrial facilities where spines are removed to allow for rapid page digitization. Once the data is extracted, the remaining paper materials are recycled, meaning the original physical copies cease to exist.
The scope of these acquisitions has left sellers bewildered by the lack of discernible pattern in the titles purchased. Manley described the orders as random, encompassing everything from obscure Latin texts to popular cowboy novels. Experts suggest this diversity supports the AI hypothesis, as large language models benefit from varied and rare datasets to improve their training. The inclusion of unusual or out-of-print works provides fresh material that may not be available through standard web scraping or commercial databases.
Anthropic has acknowledged that sourcing books for training is a common industry practice, stating that Claude is trained on a mix of public web data, commercially acquired datasets, and self-generated information. A spokesperson emphasized that the company does not target rare or antiquarian volumes for destruction. However, the confirmation that physical books are being pulped after scanning has generated unease among cultural custodians who view these items as historical artifacts rather than mere data sources.
Legal experts point out significant differences between US and UK copyright regimes regarding this activity. Professor Emily Hudson of Oxford University noted that under UK law, creating training libraries and conducting the training process generally require permission from copyright owners. This regulatory divergence highlights the complexity of cross-border data acquisition, as companies may exploit more permissive jurisdictions to gather materials that would be restricted elsewhere.
Booksellers hold conflicting views on the ethical implications of this trend. David Tobin of Walden Books in London expressed sadness at the potential destruction of titles that have remained unsold for years, even while acknowledging the financial benefit. Derek Walker of McNaughtan’s in Edinburgh raised concerns about the loss of unique historical editions, noting that some surviving copies from the 18th century are irreplaceable. He distinguished between common academic texts with multiple library holdings and singular artifacts whose destruction would represent a significant cultural loss.
Conversely, some retailers view the recycling aspect as a practical solution for oversaturated markets. Manley argued that the world does not need millions of copies of bestsellers like The Da Vinci Code, many of which have languished unsold for years. For sellers who have listed certain titles online for decades without success, the sudden demand provides an opportunity to clear inventory and generate revenue. The situation presents a complex trade-off between economic viability for small businesses and the preservation of physical literary heritage.
As the AI industry continues to expand its data requirements, the tension between technological progress and cultural preservation is likely to intensify. While the legal framework in the US currently supports these acquisition methods, the global reaction suggests ongoing debate about the value of physical books beyond their informational content. Booksellers remain uncertain about the long-term impact on their trade and the broader implications for how society values printed media in an increasingly digital age.
Sources behind this briefing
Go to the original reporting
- BBC News↗Secondhand book sales are booming. Is it because of AI?
