Elon Musk does not like the way Anthropic treats a book. In a series of posts on X this week, the Tesla and SpaceX chief executive attacked the AI company’s practice of destructive book scanning, in which volumes are cut apart to be fed through scanners for model training, and said he has ordered his own AI operations to preserve rare books instead.
The target of Musk’s criticism is an Anthropic program, known internally by a code name, that involves scanning books at industrial scale. The Washington Post has reported that Anthropic has been digitizing books from around the world since early 2024, using methods that include slicing off the spines so pages can be fed through high-speed scanners.
The practice is common in the book-scanning business, where speed matters more than preservation. Libraries and archives that want to digitize a collection quickly often cut the binding; institutions that want to preserve the physical volume use slower, more expensive equipment. Anthropic’s scale, and the fact that the books are being scanned to train AI models rather than to preserve culture, has drawn criticism from authors and publishers.
Musk’s intervention adds a prominent voice to that criticism. He said he has asked his AI company to keep the rare books it processes in a library and to scan them in ways that do not damage them, positioning the request as a contrast with Anthropic’s methods.
The exchange is the latest in a long-running feud between Musk and Anthropic’s founders. Musk was an early investor in OpenAI and has since become one of its loudest critics, and Anthropic, founded by former OpenAI researchers, sits in the same crosshairs. His posts on the book-scanning issue fold a cultural grievance into that broader conflict.
The copyright fight behind the scanning is the more consequential issue. AI companies have built their models on large collections of copyrighted books, arguing that training on published text is fair use. Publishers and authors have sued, and the cases are working their way through the courts. The way the books were physically handled has become a public-relations liability on top of the legal one.
For Anthropic, the practical problem is that its training data practices are under scrutiny from several directions at once. It faces copyright litigation, regulatory questions about data sources, and now a public campaign from the most-followed figure in American business. Each attack makes the company’s next statement on the topic more difficult.
The book preservation question has a long history in the technology industry. Google spent more than a decade defending its own book-scanning project, which used non-destructive methods, and settled the resulting litigation. The archive community has built elaborate standards for how to digitize without destroying, and those standards are available to any company that wants them.
Musk’s own position is not without complications. His companies collect vast amounts of data for their AI efforts, and critics have questioned whether SpaceX’s AI operations are any more scrupulous about the source of their training material. The distinction Musk is drawing, between how a book is treated physically and whether it is used at all, is one his own critics may probe.
The Washington Post’s reporting described the scale of the operation. Anthropic, the report said, has been scanning books in bulk since early 2024, feeding large volumes of text into the training runs that build its models. The company has said it licenses some material and negotiates with rights holders, but the volume of what it has scanned exceeds what licensing agreements could plausibly cover, critics say.
The copyright litigation provides the backdrop. Publishers and authors have sued the major AI companies over the use of their books in training, and the cases have moved through the courts without a clear resolution. Anthropic has said it respects creators’ rights and has been working toward licensing arrangements, a position that the physical handling of books complicates.
The preservation argument gives the dispute a different audience. Libraries, archives and the digitization industry have spent decades developing standards for scanning that keep the original volumes intact, and they have watched the AI industry’s methods with alarm. Musk’s posts, whatever their motive, put that technical distinction in front of a much larger public.
The feud’s history ensures the exchange will continue. Musk and the founders of Anthropic have traded public criticism for years, and the book-scanning dispute gives Musk a fresh angle. For Anthropic, the challenge is that each round of the argument, regardless of the merits, reinforces the image of a company that took shortcuts to build its models.
The books themselves remain the unresolved question. Anthropic has said it respects the rights of authors and has been negotiating licensing agreements. But the image of a rare book being cut apart for the sake of a training run is a hard one to walk back, and Musk’s posts ensure it will stay in view.
