Can AI Companies Legally Read Your Books Without Asking? Courts Are Still Figuring It Out

The basic problem

To get good at writing, chatbots like ChatGPT, Gemini, and Claude were trained by reading enormous numbers of books and articles — hundreds of millions of them. The catch is that almost none of the authors of those books ever agreed to this or even knew it was happening. Copyright law generally says you need permission to use someone's creative work, so on the surface this looks like a problem. In practice, the legal answer is a lot messier.

A landmark case, and a surprising ruling

The clearest test case so far involves Anthropic, the company behind the Claude chatbot. A judge ordered Anthropic to pay $1.5 billion to a group of authors — but here's the twist: the judge didn't rule that training AI on their books was illegal. He ruled that part was actually fine. What Anthropic got in trouble for was how it obtained some of those books — through pirate websites that illegally distribute copyrighted material, rather than buying legitimate copies.

The judge compared an AI reading books during training to a human writer who reads a lot of novels to learn their craft: the AI isn't supposed to copy those books, just learn general patterns from them, the same way an aspiring novelist might learn from studying other writers without plagiarizing them. One legal expert pointed out that a $1.5 billion fine, while it sounds huge, is actually pretty small next to how much money Anthropic is expected to make in the next few years — so this ruling isn't exactly a knockout blow against AI training.

The legal concept doing all the work: "fair use"

Almost this entire debate comes down to a legal idea called "fair use." It's a rule that says you're sometimes allowed to use someone else's copyrighted work without asking — but only if your use is different enough, or "transformative" enough, from the original. Courts look at things like: what was the new use for, how much of the original did you use, and does your new product hurt the original creator's ability to make money from their work.

A separate court case, involving a legal-research company called Ross Intelligence, showed what happens when this test fails. A judge ruled that Ross's product wasn't "transformative" because it was basically just built to compete directly with the company it copied from, doing the same job in the same way. That's a useful contrast: training on books to learn general writing skills might be seen very differently from building a product that's meant to directly replace the original source.

Why nobody has a clean answer yet

Part of the reason this is all so confusing is that the main US copyright law being used to decide these cases was written in 1976 — decades before anyone imagined AI chatbots. Judges are stuck applying old rules to a totally new kind of technology, and different courts are reaching different conclusions about whether AI chatbots that can write stories or answer questions are competing unfairly with the human authors whose work trained them.

There's also a strange follow-up wrinkle: a separate ruling found that content which is 100% generated by AI, with no human creativity involved, can't be copyrighted at all. That raises its own tricky questions about who owns what when AI is involved in the writing process, on either side of the equation.

The bottom line: authors have, without ever agreeing to it, provided a lot of the raw material that makes today's AI chatbots work — and the legal system is still in the middle of deciding what, if anything, they're owed for it.