Prompts alone do not make AI output copyrightable with current tools, the US Copyright Office found in 2025

On January 29, 2025, the US Copyright Office, based in Washington, D.C., released the second part of its report on copyright and artificial intelligence, which asks when material made with generative AI can be protected. Its answer: only where a human author has determined enough of the expressive elements. Using AI as a tool does not rule out protection, the Office found, but typing prompts does not on its own make someone the author of what a system produces. On March 2, 2026, the Supreme Court denied a petition from Stephen Thaler that challenged the Office's requirement of a human author, so the appeals court ruling that upheld it stands.

What the report concluded

The report drew on more than 10,000 comments sent in after an August 2023 notice of inquiry, about half of which dealt with copyrightability. The report concluded that existing law can resolve these questions without action from Congress, and that each work has to be judged on its own facts. The Office also concluded that nobody had made the case for extending copyright, or creating a separate sui generis right, to cover AI-generated content.

Why prompts fall short, in the Office's view

The Office treats a prompt as an instruction that conveys ideas, and ideas are not protected. Even a very detailed prompt, it argues, does not control how a system turns those ideas into fixed expression, and some systems rewrite prompts internally before using them. To show the gap, the Office entered a prompt into Google's Gemini that began "professional photo, bespectacled cat in a robe reading the Sunday newspaper and smoking a pipe." The image followed some of the instructions and not others, and the system filled in what the prompt left open, from the cat's breed and pose to a human hand holding the newspaper.

Because the same prompt can produce many different outputs, the Office sees one more sign that the user is not in control. Rewriting a prompt again and again does not change its analysis either, since copyright rewards original authorship rather than time or effort. The report leaves the door open: a future system might give users so much control that the machine's part becomes mechanical, but the Office says the evidence on today's systems shows that is not yet the case.

The arguments on the other side

Commenters disagreed on the harder cases. Most accepted that material generated entirely by AI cannot be protected, but they split over outputs that involve some human input. A few argued that revising a prompt, sometimes hundreds of times, can give a user greater control over the result. Others put forward what the report calls authorship by adoption, comparing a user who keeps prompting until the output matches the expression they want with a painter who keeps adding paint. On policy, some said protection would encourage more works and help people with disabilities to create, while others, among them representatives of copyright owners, warned that protecting AI output would discourage human authorship.

Stephen Thaler, a computer scientist, took the question to court. He applied to register an image titled A Recent Entrance to Paradise, listing his AI system, the Creativity Machine, as its sole author and himself as its owner, and the Office refused. Before the US Court of Appeals for the D.C. Circuit he argued that the natural meaning of the word author is not limited to people. On March 18, 2025, the court held that the Copyright Act of 1976 requires all eligible work "to be authored in the first instance by a human being." It added that the rule does not bar copyright for work made by or with the help of AI, as long as the author is a person.

What has happened since

The appeals court denied rehearing on May 12, 2025. Thaler filed a petition with the Supreme Court on October 9, 2025 (No. 25-449). The federal government filed its brief in opposition on January 23, 2026, and the Court denied the petition on March 2, 2026.

The report also surveyed other countries as of January 2025. Guidance in South Korea, guidelines in Japan, a 2024 court ruling in Czechia and the views of most EU member states all tie protection to human input. In 2023, the Beijing Internet Court found an image made with Stable Diffusion protected, pointing to the user's more than 150 prompts and later adjustments. The United Kingdom has a statutory provision for works generated by computer with no human author, which the Office notes predates generative AI.

What applies to registrations now

For registrations, the Office's AI page still lists the guidance it published in the Federal Register on March 16, 2023, and no later registration guidance. Under it, applicants must disclose AI-generated content in a work and briefly explain the human author's contribution, using the Standard Application. AI-generated material that is more than de minimis should be excluded from the claim, and an AI tool or the company that provides it should not be named as an author. Applicants who are unsure how to fill in the form may simply state that the work contains AI-generated material, and the Office will contact them, according to the guidance. It warns that applicants who leave the public record uncorrected risk losing the benefits of a registration.

When it released Part 2, the Office said it planned to supplement that guidance and update its Compendium of practices. Part 3, on training AI models with copyrighted works, came out as a pre-publication version on May 9, 2025. The Office's AI page says a final version will follow and that no substantive changes are expected in its analysis or conclusions.