Copyright Office says AI outputs need human creative control to be copyrightable

The US Copyright Office published Part 2 of its three-part report on AI and copyright, addressing when — if ever — output from a generative AI system can itself be copyrighted. Its conclusion was that existing law, applied case by case, already answers the question, and that no new AI-specific category of protection is needed.

The office’s central finding concerned prompting. A prompt, however detailed, does not give a user the “control over the expression” that copyright law requires of an author, because it is the system, not the person, that determines how the prompt becomes specific words, pixels or notes — so output that is wholly AI-generated cannot be copyrighted. The office distinguished this from cases where a human’s own creative expression remains discernible within an AI-assisted work: through selecting, arranging or modifying AI-generated material, or combining it with a person’s own contribution. Those human elements stay protectable, though protection extends only to them — the same logic the office had applied in 2023 when it stripped copyright from the individual AI-generated images in the graphic novel Zarya of the Dawn while keeping it for the author’s text and arrangement.

Register of Copyrights Shira Perlmutter said the conclusions rested on the centrality of human creativity to copyright’s constitutional purpose. The report drew on more than 10,000 public comments filed after the office opened its AI inquiry, and cited over 200 cases in updating the Compendium of Copyright Office practices.

The harder question — whether training AI models on copyrighted works without permission counts as fair use — was left to Part 3. That report proved contentious enough that the Trump administration fired Perlmutter days after its release, a dismissal a federal appeals court later found unlawful.