Amazon secretly buys rare books, destroys them for AI training data

Amazon's book acquisition and destruction operation represents a largely unreported dimension of AI training data sourcing. The investigation reveals that the company is not only purchasing books but systematically destroying them in the process of extracting training data—a practice that differs fundamentally from the typical understanding that AI companies license or purchase datasets while preserving original materials. The operation's secrecy (it had not been publicly reported before) and the specific infrastructure dedicated to it (the VGT3 team with its own internal branding) suggest this is an established, organized effort rather than an ad hoc data collection initiative. The targeting of rare books in particular indicates that Amazon's acquisition strategy may be indiscriminate or opportunistic, raising questions about whether the books being destroyed include materials of historical, cultural, or commercial value that publishers and authors did not consent to have used for AI training.