"Silicon Valley AI companies are starting to 'burn books' to train their models. Not literally throwing books into a fire, but in a more industrialized way: bulk-buying physical books, using hydraulic equipment to cut off the spines, scanning all the pages at high speed, and then discarding or recycling the paper copies. The company doing this is Anthropic, the team behind Claude. Court documents reveal their 'Project Panama' sought suppliers with plans to scan 500,000 to 2 million books within six months. They even brought in the former head of the Google Books project, aiming to build a digital library for model training and research. The reason is ironic: the internet is increasingly polluted with AI-generated content, making physical books written by humans—especially those published before 2022—the cleanest and rarest training data. The legal implications are even more interesting. U.S. courts ruled that as long as Anthropic legally purchases a physical book, scans it, destroys the original, and replaces it with a one-to-one digital copy, it qualifies as fair use. However, the millions of e-books they previously downloaded from piracy websites are not protected. This led to a $1.5 billion settlement, with approximately 482,000 books involved in the compensation—about $3,000 per book. In the past, humans burned books to make knowledge disappear; in the AI era, 'destroying books' is about feeding all that knowledge to machines. The real question isn’t where the paper ends up, but rather: when human-written works become scarce training resources, who has the right to buy, copy, and turn them into commercial models?" #AI #Claude #Anthropic #ProjectPanama #GoogleBooks #FairUse #DigitalLibrary