Courts are split on whether AI training on books breaks copyright

In short: The legal answer to whether AI companies can train on copyrighted books depends on the details, and courts are not fully aligned.

What's going on

AI chatbots like ChatGPT, Gemini, and Claude learn patterns from huge amounts of text, including books, articles, and papers. Many authors say their books were used without permission, and they worry these tools can hurt their income.

A key recent example involved Anthropic. A judge ordered the company to pay a $1.5 billion settlement tied to copyrighted books, but he also said the act of training the AI itself was lawful. The problem, according to the ruling, was that Anthropic allegedly got some books from illegal “shadow libraries,” which is closer to piracy than ordinary reading.

Lawyers told TechCrunch that copyright law in the US is old, with the main law dating back to 1976. Courts are trying to apply those rules to modern AI systems. Many cases turn on “fair use,” which is an exception that can allow limited use of copyrighted work without permission, such as for education, criticism, or parody. Judges look at why the material was used, how much was used, and whether it harms the original market.

One court found it was not fair use when a company trained an AI tool using Reuters content to build a competing product. That case suggests judges may be stricter when the goal is to replace the original work.

What to watch

More lawsuits are still pending, and different courts could reach different conclusions. For now, the biggest questions are whether AI training is treated more like “reading to learn” (like a writer studying books) or like copying, and whether the AI product competes directly with the original creators.

Source: TechCrunch AI