Patents

Thaler v. Vidal: Why a Machine Cannot Be a Named Inventor — Yet

The Federal Circuit held that the Patent Act's word 'individual' means a human being, so an AI system called DABUS cannot be listed as an inventor — without deciding whether AI-assisted inventions are patentable at all.

March 4, 2026
A laptop displaying neural-network diagrams beside a patent application form
Two applications named an AI system as the sole inventor; the Federal Circuit read the statute and said only humans qualify. Shutterstock
Educational content, not legal advice. This article explains general legal concepts. It does not create an attorney–client relationship. For your specific situation, consult a licensed attorney.

Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), No. 2021-2347, decided August 5, 2022, asked a question that sounds like science fiction but arrived as a clean issue of statutory interpretation: can an artificial intelligence system be named as the inventor on a U.S. patent application? Steven Thaler built and ran a system he calls DABUS — the “Device for the Autonomous Bootstrapping of Unified Science” — and filed two applications listing DABUS as the sole inventor, naming no human at all. The United States Patent and Trademark Office rejected both as incomplete for want of a valid inventor; the United States District Court for the Eastern District of Virginia (Judge Leonie M. Brinkema) granted summary judgment to the agency; and a Federal Circuit panel of Chief Judge Moore and Judges Taranto and Stark — Judge Stark writing — affirmed. The court’s answer was narrow and textual: the Patent Act defines an inventor as an “individual,” an “individual” is a natural person, and DABUS is not one. Note that this is the patent case; Thaler’s parallel loss over registering DABUS-generated artwork, Thaler v. Perlmutter, is a separate Copyright Office dispute decided under different statutory text.

At a glance

  • Case: Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), No. 2021-2347
  • Decided: August 5, 2022; panel of Moore, Taranto, and Stark, opinion by Judge Stark
  • Below: E.D. Va., No. 1:20-cv-00903-LMB-TCB (Brinkema, J.), affirmed
  • Applications: U.S. Appl. Nos. 16/524,350 and 16/524,532, each naming DABUS, an AI system, as sole inventor (a fractal-geometry beverage container and a “neural flame” attention-attracting light beacon)
  • Holding: Under the Patent Act, an “inventor” must be an “individual,” and an “individual” is a human being; an AI system cannot be a named inventor
  • Express limit: The court did not decide whether inventions made by humans with the assistance of AI are patentable, nor who would own a machine’s output
  • Status: Affirmed; certiorari denied (No. 22-919, Apr. 2023)

“Individual” means a human being

The dispositive text is short. Section 100(f) of Title 35 defines “inventor” as “the individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention,” and § 100(g) defines “joint inventor” and “coinventor” as “any 1 of the individuals who invented or discovered the subject matter of a joint invention.” The Patent Act does not separately define “individual,” so the court turned to ordinary meaning and to the Supreme Court’s own reading of the word.

The panel leaned heavily on Mohamad v. Palestinian Authority, 566 U.S. 449 (2012), in which the Supreme Court held that “individual” in the Torture Victim Protection Act “ordinarily” means “a human being, a person,” not an organization. Absent any indication that Congress meant something unusual, the same default governs the Patent Act. Thaler had no contrary statutory definition to offer and no textual signal that Congress contemplated a non-human “individual.” That was effectively the end of the case: where a statute’s text is unambiguous, a court applies it, and the court found the text unambiguous.

Two corroborating features of the statute sealed the reading. First, the Patent Act uses personal pronouns — referring to an inventor as “himself” or “herself” and never “itself” — which the court treated as confirmation that Congress had natural persons in mind. Second, § 115 requires each inventor to execute an oath or declaration attesting that “such individual believes himself or herself to be the original inventor.” An oath is a distinctively human act; it presupposes a being capable of holding and swearing to a belief. A software system cannot form or attest a belief in the sense the statute requires. These features did not create the holding so much as ratify it.

The court was careful to frame this as statutory, not metaphysical. It did not opine on whether DABUS “really” invents, whether machine creativity exists, or whether AI ought to be recognized as an inventor as a matter of policy. The question was what Congress wrote, and Congress wrote “individual.”

Inventorship as a human-only category

Thaler fits a longer line treating inventorship as personal to human beings. The Federal Circuit has previously held that corporations and sovereign States cannot be inventors — see University of Utah v. Max-Planck-Gesellschaft, 734 F.3d 1315 (Fed. Cir. 2013), which described conception as “a formation in the mind of the inventor” and inventors as natural persons even when a corporation owns the resulting rights. Thaler extends that intuition one more step: if a corporation, a legal person with rights and obligations, cannot invent, an AI system that is not a legal person plainly cannot either.

This matters because patent law sharply distinguishes inventorship from ownership. Inventors must be the humans who conceived the invention; ownership of the resulting patent can then be assigned to anyone — a company, a university, an individual investor. Thaler’s theory blurred that line. He argued, in effect, that because he owned and operated DABUS he should be able to claim its output. But the statute does not let a putative owner manufacture inventorship by attributing conception to a tool. The inventor field is not a property field; it is a factual attestation about who conceived. And the only entities the Act recognizes as capable of that conception are human.

The practical consequence is that the inventorship requirement operates as a gatekeeper. An application that names no human inventor is not merely vulnerable to challenge — it is incomplete on its face and cannot mature into a patent. Thaler’s applications failed not because the underlying ideas were unpatentable in substance, but because the applications could not satisfy a threshold administrative requirement.

The question the court refused to answer

The most consequential sentence in the opinion is a disclaimer. The court expressly stated that it was “not confronted today with the question of whether inventions made by human beings with the assistance of AI are eligible for patent protection.” That carve-out is doing enormous work. Thaler presented the hardest version of the question — an application naming a machine as the sole inventor and no human at all. The far more common real-world scenario is a human researcher who uses generative or computational tools to help produce an invention and who is, herself, a named inventor. Thaler says nothing about whether or when that human’s contribution suffices, or how much human conception is enough when an algorithm did substantial work.

By resolving the extreme case on a clean textual ground, the court avoided the genuinely difficult line-drawing that AI-assisted invention will require. That line-drawing has since migrated to the USPTO, which issued inventorship guidance in 2024 applying a Pannu-style “significant contribution” analysis to AI-assisted inventions — requiring that a named human have made a meaningful contribution to conception. Thaler is the doctrinal floor beneath that guidance: it establishes that the inventor must be human, leaving the agency and future courts to decide how much human input the law demands.

Open questions

  • How much human contribution is enough? If a person prompts, curates, and refines an AI system’s output, when does her contribution rise to the level of conception that the statute requires? Thaler does not say.
  • Who owns a machine’s unattributable output? If no human can be named for a given AI-generated advance, the invention may simply be unpatentable in the United States — a gap that Thaler leaves open rather than fills.
  • Is this a job for Congress? The court repeatedly framed its holding as compelled by current text. A statute written before modern generative AI may not reflect today’s policy choices, and only Congress can change “individual.”
  • Will jurisdictions diverge? With the U.S., the UK Supreme Court, and the EPO all rejecting AI inventorship while a few jurisdictions have been more permissive, applicants face an increasingly fractured global map for the same underlying technology.

Implications

  • Name a human, always. Every U.S. application must identify at least one natural-person inventor who actually contributed to conception; an AI tool cannot fill that field, and an application that names only a machine is incomplete.
  • Document the human contribution. Where AI tools materially assist invention, build a contemporaneous record of what specific humans conceived and contributed — that record is now the crux of inventorship validity.
  • Distinguish inventorship from ownership. Owning or operating an AI system does not make you (or it) an inventor; assignment law handles ownership, but only after a human inventor is correctly named.
  • Watch the agency, not just the courts. The live battleground for AI-assisted invention is USPTO guidance and its application, since Thaler deliberately left the harder questions unresolved.
  • Plan for a fractured world. Global filing strategy should account for divergent national rules on whether and how AI involvement must be disclosed or named.

Frequently asked questions

Did the court decide that AI-generated inventions can never be patented? No. It decided only that an AI system cannot be a named inventor. The court expressly declined to address whether inventions made by humans with the assistance of AI are eligible for patent protection.

Is this the same as the DABUS copyright case? No. Thaler v. Vidal is the patent case, decided under the Patent Act’s definition of “inventor.” Thaler’s separate copyright dispute over a DABUS-generated image, Thaler v. Perlmutter, turns on different statutory text and the Copyright Office’s human-authorship requirement.

Why did the personal pronouns and the oath requirement matter? They corroborated the textual reading. The Patent Act refers to an inventor as “himself” or “herself,” and § 115 requires each inventor to swear an oath of original inventorship — features the court read as confirming that Congress meant human beings, because a machine can neither be a “himself”/“herself” nor swear a belief.

Authorities and sources