Imagine buying a book, scanning every page into a computer and then throwing the physical book away. Sounds bizarre, right? That’s the story now surrounding Anthropic, the company behind Claude, and its efforts to build huge libraries of books for AI training.
But there’s an important detail the viral headlines often miss.
In 2025, a US court found that training Claude on books could qualify as fair use, even when the books were copyrighted. The court also found that Anthropic had illegally obtained and stored millions of pirated books. Anthropic later agreed to a $1.5 billion settlement, approved in July 2026.
Now, newly reported documents and investigations have brought another part of the story into focus: Anthropic purchased physical books, converted them into digital files, and discarded some of the physical copies after scanning. That has triggered a fresh debate about whether valuable books, especially rare ones, should be treated simply as data.
Here’s the interesting question:
If a human reads 10,000 books and learns from them, we usually call that education. If an AI does something similar, is it the same thing?
Before deciding whether Anthropic is right or wrong, ask yourself:
- Who should own the knowledge inside a book?
- Should creators be paid when their work helps build AI?
- And if a book can be digitised without destroying it, why destroy the physical copy at all?
- Would you be comfortable with an AI learning from your favourite book without asking you first? Why or why not?
Let us know your response in the Support chat. We will reach out to you for your story. 👀💛