Authorship Without an Author: Thaler v. Perlmutter and the Human Floor of Copyright

The D.C. Circuit held that the Copyright Act requires a human author, foreclosing registration of a work generated autonomously by an AI system — and the Supreme Court has now declined to disturb that conclusion.

In Thaler v. Perlmutter, No. 23-5233 (D.C. Cir. Mar. 18, 2025), the United States Court of Appeals for the District of Columbia Circuit confronted a deceptively narrow question with expansive stakes: whether a work generated autonomously by a machine, with no human creative contribution claimed, can be registered for copyright. Writing for the panel, Circuit Judge Patricia Millett answered no. The Copyright Act of 1976, the court held, requires that a copyrightable work be authored “in the first instance” by a human being. The decision affirmed the District Court for the District of Columbia (Howell, J.) and ratified what the Copyright Office has maintained for decades. On March 2, 2026, the Supreme Court denied certiorari (No. 25-449), leaving the human-authorship requirement intact as governing law in the circuit that hears most appeals from federal agency action.

At a glance

Case: Thaler v. Perlmutter, No. 23-5233 (D.C. Cir.)
Decided: March 18, 2025; cert. denied March 2, 2026 (No. 25-449)
Holding: The Copyright Act requires a human author; a work generated entirely by an AI system, with no human listed as author, is ineligible for registration.
Status: Affirmed below; certiorari denied. Final as to the question presented.

Procedural posture

The dispute traces to a registration application Stephen Thaler, a computer scientist, filed for a two-dimensional visual work titled “A Recent Entrance to Paradise.” The application was unusual on its face: Thaler listed his AI system, the “Creativity Machine,” as the sole author of the work, and named himself as the copyright claimant by virtue of his ownership of the machine. He represented that the work had been “autonomously created” by the system and disclaimed any human creative input.

The Copyright Office refused registration in 2019, and twice more on administrative review, on the ground that the work lacked the human authorship the Office considers indispensable. Thaler sued under the Administrative Procedure Act, contending that the refusal was arbitrary and contrary to law. On August 18, 2023, Judge Beryl Howell of the District Court for the District of Columbia granted summary judgment to the Office, holding that “human authorship is a bedrock requirement of copyright” and that it is “the sine qua non at the core of copyrightability.” Thaler appealed to the D.C. Circuit.

A procedural feature of the case sharply limited its reach. Thaler had, at the registration stage, insisted that the machine alone was the author. He attempted on appeal to recast the theory — arguing that he was the author because he designed, built, and operated the Creativity Machine. The D.C. Circuit declined to reach that argument, treating it as waived because it had not been presented to the Copyright Office and was not the factual predicate on which the registration was sought. The court thus decided the case on the terms Thaler himself had set: a work with no human author at all.

The court’s reasoning

The statutory text does not define “author.” Judge Millett’s opinion nonetheless found the human-authorship requirement embedded throughout the 1976 Act, reasoning that the term must be read in light of the statutory scheme “taken as a whole.” The court canvassed provisions that presuppose a human author and would be incoherent if applied to a machine. Copyright duration is measured by “the life of the author” plus seventy years (17 U.S.C. § 302(a)) — a metric meaningless for a system with no lifespan. Ownership vests initially in the author and passes by inheritance and through intestacy and community-property regimes, all of which contemplate a natural person. The Act’s joint-authorship definition turns on the authors’ intention to merge their contributions (§ 101), and termination-of-transfer rights run to the author’s surviving spouse and children (§ 203). Machines do not form intentions, take spouses, or leave heirs.

The court also leaned on interpretive history. It found that the Copyright Office had consistently construed “author” to mean a human being before Congress enacted the 1976 Act, and that Congress, in re-enacting the term against that settled administrative backdrop, is presumed to have adopted the established construction. This is a familiar move in statutory interpretation — the ratification canon — and it allowed the court to ground the human-authorship rule in legislative intent rather than abstract policy.

Critically, the court resolved the case on statutory grounds and did not rest on the Constitution. It acknowledged but did not decide whether the Intellectual Property Clause’s reference to “Authors” independently requires humanity. By keeping the holding statutory, the panel left Congress room to legislate and avoided constitutionalizing a question that generative AI is still rapidly reshaping.

Doctrinal context

The decision sits comfortably within a long line of authority. In Burrow-Giles Lithographic Co. v. Sarony, 111 U.S. 53 (1884), the Supreme Court upheld copyright in a photograph of Oscar Wilde against the argument that a camera, not a person, made the image — locating authorship in the photographer’s creative choices of pose, lighting, and arrangement. Burrow-Giles is often misread as a charter for protecting machine output; it is the opposite. It protects the human who directs the instrument, and its reasoning presupposes a human mind behind the apparatus. The Ninth Circuit’s Naruto v. Slater, 888 F.3d 418 (9th Cir. 2018) — the “monkey selfie” case — reinforced the point from the other direction, holding that a non-human (there, a macaque) lacks statutory standing to claim authorship. Thaler extends that logic from animals to algorithms.

The Copyright Office’s contemporaneous guidance frames the practical landscape. Its 2023 statement of policy on works containing AI-generated material, and the registration-and-authorship analysis in its multi-part report on copyright and artificial intelligence, both maintain that purely machine-generated output is unprotectable while works reflecting sufficient human authorship — through selection, arrangement, or creative modification of AI output — may be registered, with the AI-generated elements disclaimed. Thaler validates the floor of that framework without mapping its contours.

Open questions

The opinion is studiously narrow, and its silences matter more than its holding for most practitioners.

The central unresolved issue is the threshold of human contribution. Thaler tells us that zero human authorship yields no copyright; it says nothing about how much is enough. Does iterative, detailed prompting constitute authorship? Does curating and arranging a hundred AI outputs into a composite work? Does substantial post-generation editing? The court expressly reserved “the thornier and more fact-specific questions regarding the degree to which human creativity reflected in AI-generated works may be copyrightable.”

A second question is the durability of the “autonomous creation” predicate. Thaler’s case was unusually clean — and unusually weak — because he affirmatively disclaimed human input. Most real-world disputes will involve hybrid works where the human-versus-machine contribution is contested and difficult to disaggregate.

Third, the constitutional question remains genuinely open. Because the panel decided the case on the statute, a future litigant could argue that Congress may extend protection to machine output, or, conversely, that the Constitution forbids it. The cert denial settles the statutory point for now but forecloses neither legislative reform nor a future constitutional test.

Implications

For creators and studios: Workflows that rely heavily on generative tools should document human creative decisions — selection, arrangement, editing — and be prepared to disclaim purely machine-generated elements when registering.
For litigators: Registration applications are now strategic documents. Thaler’s defeat was sealed at the administrative stage by how he characterized authorship; the waiver ruling is a cautionary tale about preserving theories before the agency.
For the Copyright Office: The decision is a substantial vindication of its longstanding position and its recent AI guidance, strengthening its hand in examining hybrid applications.
For licensing and transactions: Counsel should assume that wholly AI-generated assets may fall into the public domain, affecting representations, warranties, and indemnities in content deals.
For policymakers: By resting on statute, the court left the question of whether to protect machine output squarely with Congress.

Frequently asked questions

Does Thaler mean nothing made with AI can be copyrighted? No. The holding addresses works with no claimed human author. Works reflecting meaningful human creative contribution may still be registered, with the AI-generated portions disclaimed. The court deliberately left the human-contribution threshold for another day.

Why didn’t Thaler simply claim he was the author? He tried, but only on appeal. Because he had told the Copyright Office the machine was the sole author and disclaimed human input, the D.C. Circuit treated the contrary argument as waived and decided the case on the record he created.

Is the question now settled? As to fully autonomous machine output, effectively yes within the D.C. Circuit, and the Supreme Court’s March 2026 cert denial leaves the ruling undisturbed. But cert denials are not merits rulings, the human-contribution line remains unlitigated, and Congress retains authority to change the statute.