In 2022, James Allen filed a copyright application for an image titled “Théâtre D’opéra Spatial,” or “Space Opera Theater.” While the picture had the look of a detailed sci-fi painting, Allen had actually created it through painstaking experimentation in the AI image generator Midjourney; he described using over 600 prompt variations to get the look he wanted. But the US Copyright Office wasn’t impressed. In a series of decisions, Allen’s work became one of the first copyright requests rejected specifically for using AI tools — and an example of how these tools are raising new copyright conundrums at virtually every turn.

Generative AI programs — particularly image generators like Midjourney, Stable Diffusion, and DALL-E — have raised fundamental questions about how far artists’ rights to their work extend. Many are trained on huge datasets that include copyrighted works without artists’ consent, and in the US, a series of lawsuits could determine whether the applications fall under the framework of exceptions known as fair use. Meanwhile, the US Copyright Office holds that a computer program can’t make copyright-protected art, frustrating people who see DALL-E or Stable Diffusion as tools akin to Photoshop.

Copyright exists to encourage the production of art, but copyright law also recognizes that most art draws on work that came before it — that’s why exceptions like fair use exist. There’s a constant balancing act in play, and when a new technology appears, it can require renegotiating that balance. Where will it fall in the case of AI? So far, nobody knows for sure.