Artificial intelligence systems such as ChatGPT, Gemini and Claude are being trained on vast collections of books, articles and research papers, placing copyright law at the center of a fast-moving technological debate.
According to intellectual property attorney Cathy Gellis, the issue is far more nuanced than a simple yes-or-no answer. Courts are now weighing whether training an AI model is closer to reading a work for learning or copying it in a way that harms the original creator.
A recent ruling involving Anthropic highlighted that distinction. Judge William Alsup found the company's model training itself lawful, while penalizing the use of books obtained from illegal digital libraries. The decision suggested that the method of access can matter as much as the training process itself.
Legal experts say the debate now turns on fair use, a doctrine that allows limited use of copyrighted material for purposes such as education, commentary and transformation. Courts are examining whether AI training changes a work enough to qualify as transformative, or whether it competes too directly with the original market.
Another key question is how copyright applies to content produced by AI. In one case, a court ruled that fully AI-generated works are not eligible for copyright protection, adding a new layer to the discussion about authorship, originality and creative control.
Because U.S. copyright law has not been substantially updated since 1976, judges are interpreting older rules for a new era of machine learning. For now, the legal landscape remains unsettled, but each ruling is helping define the boundaries of AI development and creative rights. The outcome could shape how future tools are built, licensed and trusted across the digital economy.
