Amazon and Anthropic Uncovered Destroying Physical Books for AI Training Data
An investigation by 404 Media using Apple AirTags revealed that Amazon is purchasing physical books, cutting off their spines for rapid scanning to train its Nova AI models, and destroying the originals. A similar destructive scanning initiative, known as "Project Panama," was previously conducted by Anthropic. This practice highlights the extreme measures AI companies take to acquire high-quality, pre-2022 training data that is free of AI-generated content. However, it raises significant ethical concerns as rare, irreplaceable books are permanently lost to the public and locked inside proprietary corporate models. A court previously ruled that Anthropic's scanning constituted fair use, partly because the physical books were destroyed, preventing the resale of duplicate copies. The targeted books are highly valued by AI developers because their contents have never been published on the internet.
## BACKGROUND
AI developers face a growing shortage of high-quality, human-written text to train large language models, leading them to seek offline sources like physical libraries. Amazon's Nova is a family of foundation models launched on AWS, while Anthropic's "Project Panama" was a controversial data-harvesting operation that faced legal action from authors.