AI Regulation Tracker  /  Copyright, authorship and AI-generated work

A Federal Appeals Court Says Your AI Cannot Be the Author

In Thaler v. Perlmutter, the U.S. Court of Appeals for the D.C. Circuit ruled on March 18, 2025 that a work generated autonomously by artificial intelligence, with no human author, cannot be registered for copyright. The court held that the Copyright Act of 1976 requires every eligible work to be authored in the first instance by a human being. This is not the newest development, but it is now settled: in March 2026 the Supreme Court declined to hear the case, so the human-authorship rule is controlling federal appellate precedent. If your firm wants to own and register what its AI produces, this decision tells you exactly where the line sits.

By Anthony Guerriero  ·  The Leveraged Years Regulatory Research  ·  Published July 20, 2026  ·  Last updated July 20, 2026

Let me be straight about the date, because it changes how you use this. The decision came down on March 18, 2025, and the underlying application goes back to 2019. What makes it worth a place in the tracker today is that it is now settled at the top. In March 2026 the Supreme Court denied review, which means the D.C. Circuit opinion is the controlling appellate authority and there is no higher court waiting to change it. If you are deciding right now whether your firm can register what its models produce, this is the case that governs the answer.

The facts are clean, and that is the point. Thaler did not claim he was the author. He listed the Creativity Machine as the sole author and wrote on the application that the work was "Created autonomously by machine." He repeated to the Copyright Office that the submission "lacks traditional human authorship." So the court never had to draw a hard line about how much human input is enough. It had a work with a human author of exactly zero, and it ruled on that.

The holding: author means human

The court stated the rule at the top of its analysis. "As a matter of statutory law, the Copyright Act requires all work to be authored in the first instance by a human being." The Copyright Act does not actually define the word author, so the court read the statute as a whole and walked through provision after provision that only makes sense if an author is a person. Copyright vests in the author immediately, which assumes the author can hold property. Protection runs for the life of the author plus seventy years, which assumes a human lifespan. Rights pass to a widow, widower, children, or grandchildren. Transfers require a signature. Authors have a nationality and a domicile, and joint authors have an intention to merge their contributions. Machines, as the court put it, have none of these things.

From that structure the court drew the conclusion that the best reading of the statute makes humanity a necessary condition for authorship. It reinforced the point with history, noting that the Copyright Office has required a human author since at least 1973, that a congressional commission studying computers and copyright in the 1970s concluded a computer is "an inert instrument, capable of functioning only when activated either directly or indirectly by a human," and that Congress has never amended the Act to allow machine authorship.

What the decision does not do

This is the part professionals most often get wrong, so read it carefully. The court did not rule that AI-assisted work is uncopyrightable. It said the opposite. In its words, "the human authorship requirement does not prohibit copyrighting work that was made by or with the assistance of artificial intelligence. The rule requires only that the author of that work be a human being." The court pointed out that the Copyright Office already registers works where a human uses AI as a tool. What is barred is naming the machine itself as the author of a work with no human creative contribution.

The court also left two questions open on purpose. It did not decide whether the Constitution independently requires human authorship, because the statute was enough. And it did not decide Thaler's argument that he was the author by virtue of building and running the machine, because he had waived that argument before the agency. So this case does not tell you how much human involvement converts an AI output into a registrable human-authored work. It tells you that a pure machine output, with the machine named as author, is out.

Why this is a US practice problem, not a trivia item

Here is the hook, and it is why an attorney should care beyond the novelty. Copyright ownership is the hinge for a lot of commercial value. If your firm generates marketing copy, code, images, or reports with a generative model and there is no meaningful human authorship, you may not be able to register that output, and registration is a precondition for suing to enforce a US copyright. That affects what you can license, what you can claim exclusivity over, what you can put on a balance sheet as a protectable asset, and what you should promise a client in an engagement letter or a work-product warranty.

The defensible posture after Thaler is documentation of the human hand. The registrable version of AI-assisted work is the one where a person selected, arranged, edited, and shaped the output enough to be a real author, and where you can show that contribution. The unregistrable version is the one you let the model produce end to end and then tried to claim as your own. For attorneys, the practical work is advising clients to build that human-authorship record into their process now, and to stop assuming that whatever the machine spits out is automatically the firm's to own and register.

Browse the full AI Regulation News tracker

Informational analysis for working professionals, not legal advice. Confirm how Thaler v. Perlmutter and the human-authorship requirement apply to your firm's AI-generated work, ownership, and registration strategy with qualified copyright counsel.