The rapid growth of generative AI has created a major legal question: Can AI companies train their models using copyrighted books without permission from authors and publishers?

According to a recent TechCrunch analysis, there is still no simple answer. In the United States, courts are increasingly examining AI training under the concept of fair use, which can allow copyrighted material to be used without permission in certain circumstances.

One of the most important cases involves Anthropic, the company behind Claude. A U.S. judge previously found that using books to train an AI model could qualify as fair use when the training creates something new rather than simply reproducing the original works. However, Anthropic also obtained millions of books from unauthorized “shadow libraries.” That part of the case became a major legal problem.

Anthropic ultimately agreed to a $1.5 billion settlement with authors and publishers over the books obtained from those sources. The case demonstrated an important distinction: AI training may potentially be legally defensible, but obtaining copyrighted material illegally can create enormous liability.

Another major issue is market competition. Courts are considering whether AI-generated content directly competes with the original books and whether it causes measurable harm to authors and publishers. Different judges have reached different conclusions, showing that existing copyright law was not designed for today’s AI systems.

For businesses developing or using AI, the lesson is becoming clearer: where training data comes from matters just as much as how the AI model uses it. Companies need to pay closer attention to data licensing, provenance, copyright compliance, and the potential commercial impact of AI-generated content.

The legal battle is far from over, and future court decisions — and potentially new legislation — will determine how AI and copyright coexist.

Source of news: https://techcrunch.com/