The legality of AI training on copyrighted works is one of the most important unresolved questions in modern copyright law. It affects the publishing industry broadly, but primarily involves lawsuits against AI companies — not individual authors.

The core issue is whether training on copyrighted material is fair use. AI models are trained on large datasets that include books, articles, and other creative works. Rights holders argue this is unauthorized copying, while AI companies argue that training learns patterns rather than reproducing specific works.

Several major lawsuits are currently underway:

• New York Times v. OpenAI: Claims unauthorized use of journalism and potential reproduction of content
• Authors Guild v. OpenAI and Meta: Class action alleging unlawful use of books for training
• Additional cases from artists, musicians, and publishers

The U.S. Copyright Office has weighed in. In its 2025 report, it concluded that AI training is not automatically fair use and must be evaluated case by case. Training on unauthorized or pirated content weighs against fair use.

What this means for authors using AI:

• You are not currently liable for how AI models were trained
• Your responsibility is the content you publish, not the training data behind the tool

Future rulings could have indirect effects:

• AI tools may become more expensive if licensing is required
• Model capabilities could change if training data is restricted
• Some systems may shift to licensed-only datasets

For now, no action is required from authors. The legal battles are ongoing, and outcomes will shape the ecosystem — but not your current ability to use AI tools.

Sources:

• U.S. Copyright Office: AI Report Part 3 (Training)
• New York Times Co. v. OpenAI Overview
• Authors Guild v. OpenAI
• Congressional Research Service: Generative AI and Copyright Law