• An investigation found AI companies, including Amazon and Anthropic, are reportedly destructively scanning physical books to create training datasets for AI models.
• Anthropic’s initiative, reportedly called “Project Panama,” involved scanning books at scale.
• The practice has raised concerns over copyright, cultural preservation and data ethics, drawing criticism from authors, booksellers, academics and readers.

• Demand for high-quality data: AI models need vast, original, high-quality datasets as the internet grows saturated with AI-generated content.
• Value of rare/out-of-print books: They offer unique, untapped textual data unavailable elsewhere online.
• Competitive advantage: The company that sources a rare book first can destroy the original, denying rivals the same text — a form of data monopoly.

Why Destroy Rather Than Preserve

• Cost-effectiveness: Destructive scanning is cheaper and faster than non-destructive methods.
• Legal insulation: Buying and physically destroying a book avoids piracy accusations — a buyer may legally destroy a purchased book, unlike downloading pirated e-books.

• Fair Use: A US court held that converting a legally purchased book into a digital copy for private use can qualify as fair use.
• Different cases: The Internet Archive and shadow libraries faced lawsuits for making copyrighted books freely available online, showing the legal uncertainty around AI training and copyright.

• Loss of cultural heritage: Destroying rare, irreplaceable books may erase undiscovered works of literature forever.
• Public exclusion: Digital copies created are not made publicly accessible, unlike preservation platforms — knowledge is privatised, not democratised.
• Transparency deficit: AI model development is opaque by design, limiting accountability.