ทันเอไอ

AI companies use “destructive book scanning” for model training data to avoid content contamination

An investigative report reveals that several AI companies are cutting bindings and separating pages of real books for scanning to prevent AI degradation from synthetic data.

The AI industry is facing new ethical questions following confirmed reports that several AI companies are using a method called destructive scanning—cutting book spines and separating pages to feed into high-speed scanners, which destroys the books in the process.

The reason behind destroying large numbers of books stems from the demand for high-quality, genuinely human-written training data from before 2022 that hasn't been contaminated by AI-generated content. This is to prevent Model Collapse, a condition where AI performance degrades after learning from data generated by other AIs.

This issue came to light through an investigation by *404 Media*, which tracked used books embedded with AirTags and found they ended up at an Amazon facility in Las Vegas where books are destroyed for scanning. Additionally, court documents in an *Anthropic* lawsuit revealed a project called "Project Panama," reflecting similar efforts, along with legal debates over whether buying and destroying books for scanning falls under Fair Use.

In response, consumer rights groups and public organizations have begun taking action by filing a petition with the U.S. Federal Trade Commission (FTC) to investigate the destruction of cultural resources and data monopolization in the industry.