Copyright Office concludes that prompts alone do not make the user the author of what a model generates

2025-01-29Copyright

United States Copyright Office, "Copyright and Artificial Intelligence, Part 2: Copyrightability," A Report of the Register of Copyrights (January 2025), released January 29, 2025

The Copyright Office released the second part of its report on copyright and artificial intelligence, on whether material produced with generative systems can be protected. Drawing on more than 10,000 comments, the Office concluded that existing law answers the question without legislative change. Using artificial intelligence as a tool to assist human creation does not limit protection, and including generated material in a larger human work does not defeat protection of the human contribution, but material whose expressive elements are determined by the machine is not protected. On prompts, the Office stated that "prompts alone do not provide sufficient human control" to make the user the author of the output, given the functioning of currently available technology. A person whose own expressive input remains perceptible in the output is the author of at least that portion, and a person may claim copyright in the creative selection, coordination or arrangement of generated material and in creative modifications of it. Whether enough human contribution exists is decided case by case. The Office found that the case has not been made for additional copyright or sui generis protection for generated material. Part 1, on digital replicas, appeared in July 2024; a third part, on training, is to follow, and the Office plans to update its March 2023 registration guidance and the Compendium of its practices.

What it changesThe Office's line between assisting a human creator and replacing one is the line the patent office drew for inventors in February 2024, when its guidance said that a person who only presents a problem to a model may not be an inventor. A company now faces one documentation question under both regimes: which person made the decisions that appear in the protected result. For code, documentation, interface designs and images produced with generative tools, the protectable part is the human selection, arrangement and modification, and it is protectable only if the company can show what was done; capture the human edits in version control rather than relying on prompt history. For inventions, the same records support naming the right inventors. Raw output of a prompt carries no copyright, so a product whose value lies in generated content needs patents on the system that produces it, trade secret protection for the prompts and pipelines, and contracts that allocate rights in the outputs.