Anthropic and Other AI Firms Scan and Destroy Millions of Old Books
Major artificial‑intelligence companies are buying large volumes of printed books published before 2022, seeking “clean” human‑written text for training large language models. The books are supplied in bulk by data‑broker ISBNdb and other intermediaries, then processed with high‑speed “destructive scanning” that cuts off the spine, feeds pages through industrial scanners and discards the original copies.
Anthropic’s internal programme, dubbed “Project Panama,” aimed to acquire and scan up to two million titles, a plan that has drawn particular scrutiny. A U.S. federal judge ruled that scanning legally purchased books can qualify as fair use, even when the physical copies are destroyed, while the use of pirated copies does not. Anthropic later settled related lawsuits for $1.5 billion.
Critics warn that the practice threatens rare, out‑of‑print works and could cause irreversible cultural loss, as many of the books destroyed are the only surviving copies of certain titles. The controversy has sparked a global debate over the balance between AI development and preservation of physical literary heritage.
Entities
AI companies · Anthropic · Elon Musk · ISBNdb · Project Panama · Rare books · SpaceXAI · U.S. federal court · U.S. federal judge · Zoom Books
Claims
What the coverage asserts, and how well corroborated each claim is across sources.
- [● 4 SOURCES] The practice raises copyright concerns and is part of a broader debate on AI data sourcing. www.perlentaucher.de · businessmag.al · crypto-times.jp · technews.tw
- [● 3 SOURCES] Anthropic specifically purchased many books, as revealed in court documents. www.perlentaucher.de · crypto-times.jp · technews.tw
- [● 6 SOURCES] The purchased books are dismantled and scanned using industrial equipment. businessmag.al · www.perlentaucher.de · crypto-times.jp · technews.tw · decrypt.co
- [● 6 SOURCES] AI companies are buying large quantities of used physical books to train large language models. businessmag.al · www.perlentaucher.de · crypto-times.jp · technews.tw · decrypt.co
- [● 4 SOURCES] After digitisation, the physical copies are discarded or destroyed. businessmag.al · www.perlentaucher.de · crypto-times.jp · technews.tw
- [● 3 SOURCES] Companies argue the practice falls under fair‑use for research and educational purposes. www.perlentaucher.de · crypto-times.jp · technews.tw
- [● 4 SOURCES] The practice is motivated by the need for clean data not influenced by AI‑generated text. businessmag.al · crypto-times.jp · technews.tw · www.perlentaucher.de
- [● 4 SOURCES] European antiquarian dealers have seen a surge in large orders up to €50,000, mainly for scientific literature. www.perlentaucher.de · technews.tw · crypto-times.jp · decrypt.co
- [● 5 SOURCES] A U.S. federal judge ruled that scanning legally purchased books into digital copies, even when the originals are destroyed, constitutes fair use. crypto-times.jp · technews.tw · www.perlentaucher.de · decrypt.co
- [● 4 SOURCES] ISBNdb is offering AI companies the procurement of up to one million physical book titles for training data. crypto-times.jp · technews.tw · www.perlentaucher.de · decrypt.co
- [○ 1 SOURCE] Anthropic previously pirated more than seven million books from LibGen and Books3 before switching to purchases.
- [● 2 SOURCES] Anthropic ran an internal project called “Project Panama” that bought, scanned and shredded millions of books. decrypt.co