Anthropic Book Destruction Raises Rare Book Loss Concerns, Musk Responds
The destructive scanning method, chosen for cost and speed, risks losing books that survived wars and centuries, a trade-off that even a rival AI leader publicly rejected, signaling a potential industry split on training data ethics.
Key Facts
- Anthropic cut the bindings off millions of printed books, scanned them for AI training, and discarded the originals, according to court documents from a settlement approved on July 27, 2026.
- Anthropic hired Tom Turvey, former head of partnerships for Google Books, in February 2024 with instructions to obtain every book in the world.
- The used book market has seen buyers believed to be from IT companies purchasing large quantities of specialized and out-of-print titles, according to Dutch media NL Times.
- Elon Musk said he asked the SpaceXAI team to preserve rare books in libraries and scan them carefully rather than cutting spines and discarding them.
Reporting from 1 source: GIGAZINE.
Court documents from a settled lawsuit reveal that Anthropic cut the bindings off millions of printed books, scanned them for AI training data, and discarded the originals, a method that has drawn criticism from rare book dealers and others concerned about the loss of valuable works. The documents, approved as part of a July 27, 2026 settlement between Anthropic and a group of authors over the use of pirated books, show that the company hired Tom Turvey, former head of partnerships for Google Books, in February 2024 and instructed him to obtain every book in the world. Anthropic's approach contrasts with Google's non-destructive scanning of library books, which are returned after digitization. Instead, Anthropic bought used books in bulk, cut off spines to separate pages, scanned them with high-speed scanners, and used OCR to convert them to text, discarding the paper originals. The used book market has seen buyers believed to be from IT companies purchasing large quantities of specialized and out-of-print titles, according to Dutch media NL Times. Some dealers worry that first editions and other rare editions are being destroyed, while others see increased sales. In response, Elon Musk, who leads AI company SpaceXAI, said he asked his team to preserve rare books in libraries and scan them carefully rather than cutting spines and discarding them.
Anthropic's shift from pirated e-books to paper books followed a 2025 ruling that AI companies do not need authors' permission to train on legally obtained books, but the use of pirated data was questioned. The company aimed to replicate Google's legal victory in a 2015 fair use case, but chose a different method. ISBNdb, which supplies bibliographic data to AI companies, notes that print books published before 2022 are ideal training data because they lack AI-generated text, avoiding the 'AI slop' cycle.
The court documents did not mention that valuable books were destroyed, but rare book dealers expressed concern. A used book dealer told 404 Media that books that survived wars, fires, and centuries of handling are being discarded through destructive scanning. Anthropic's method is legally considered fair use, but critics point out that Google and the Internet Archive use non-destructive scanning.
Musk's comment, made in response to the reports, said, 'I asked the SpaceXAI team to preserve rare books in libraries and scan them carefully rather than simply cutting off the spines and scanning them.' The statement positions SpaceXAI against Anthropic's approach, though it does not detail any specific policy.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.