Anna’s Archive is asking volunteers worldwide to scan and upload physical books and other materials, arguing that AI demand for printed works is creating a preservation risk. The appeal came in a guest post by a volunteer identified as “u,” according to the Anna’s Archive post surfaced on Hacker News and covered by Tom’s Hardware. The core claim is narrower than the rhetoric around it: Anna’s Archive says AI companies are buying large quantities of secondhand books, scanning them for training data, and destroying the physical copies in the process. Tom’s Hardware reports that the appeal follows multiple reports of large AI companies buying books to train models, with some workflows resulting in the destruction of the original works. The preservation concern centers on books published before the recent surge of machine-generated text. The Anna’s Archive post describes these works as valuable because they are “untouched by machines” before 2022, a framing Tom’s Hardware also notes. The argument is that older print material can be attractive training data precisely because it predates the contamination problem created when models train on model-generated output. Tom’s Hardware adds a concrete example from the book supply chain: independent bookstores in Europe have reportedly seen unusual large orders, including obscure or dated titles that would not typically draw demand. The outlet says an investigation found that some books were sent to an Amazon processing facility, where workers removed spines and fed pages into industrial scanners. That process is faster than preservation scanning, but it destroys the book. The Anna’s Archive post also points to Anthropic, alleging that a confidential “Project Panama” began in early 2024 and involved buying millions of paper books, scanning them for Claude, and destroying them. That specific project account is attributed to the Anna’s Archive post in the provided material. Separately, Tom’s Hardware reports that Anthropic reached a $1.5 billion copyright settlement, paying $200 per title based on a collection of 7 million pirated books, and says the same ruling affirmed fair use for training AI models on existing works. Anna’s Archive’s proposed response is mass volunteer scanning. The post asks people to scan books, journal articles, newspapers, magazines, ancient books, rare books, and other materials from libraries and archives, especially items that could be easily lost. Its stated goal is to make preservation copies available through the shadow library before scarce physical works vanish into private training pipelines or are damaged by destructive digitization. The evidence here is still uneven. The call for volunteers is clear, and Tom’s Hardware reports related book-buying and destructive-scanning claims. But the provided cluster does not independently confirm every allegation about named AI companies, exact volumes, or internal projects. For readers, the defensible takeaway is that AI training data demand is now reaching into the physical book market, and preservation advocates are responding by trying to digitize vulnerable works first. Who benefits: AI companies with access to large private scans may benefit if scarce print material becomes proprietary training data. Preservation projects such as Anna’s Archive benefit if the appeal brings in volunteers and more scans. Who's exposed: Rare-book owners, small bookstores, and libraries are exposed if demand shifts toward bulk acquisition and destructive scanning. AI firms are exposed to scrutiny over how they source, scan, retain, and document training materials.