Can AI Be an Inventor? What Patent Law Says in 2026

One of the most-asked questions in AI patenting has a settled legal answer: can AI be an inventor on a patent? No. In the United States, the United Kingdom and at the European Patent Office, an inventor must be a natural person, and an AI system — however autonomously it generates an idea — cannot be named. That single rule shapes how every AI-assisted invention is filed, and getting it wrong can invalidate a patent. This guide walks through the cases that decided it, the USPTO’s critical 2025 reversal, and what it means for anyone reading an AI patent landscape and planning where to file.

Can AI Be an Inventor Under Today’s Patent Law?

The direct answer to can AI be an inventor is no, in every major patent system. The US Patent Act defines an inventor as “the individual . . . who invented or discovered the subject matter” (35 U.S.C. § 100(f)), and courts have read “individual” to mean a human being. The same conclusion holds at the UK Intellectual Property Office and the European Patent Office.

This is not a technicality that clever drafting can dodge. Inventorship is a substantive requirement: a US patent that names the wrong inventor — or names a machine — can be held invalid or unenforceable. So the question is not academic for anyone whose R&D now runs through machine-learning tools; it decides whether the resulting patent stands up.

The reason the question keeps coming up is that generative models genuinely contribute to inventions now. But the law draws a clean line: an AI system can be the instrument of an invention without being its inventor, exactly as a microscope or a simulation package can. What matters legally is the human contribution behind the filing, not the sophistication of the tool.

Thaler and DABUS: The Cases That Settled It

The definitive test case is Thaler v. Vidal. Dr. Stephen Thaler filed two US applications naming an AI system called DABUS (“Device for the Autonomous Bootstrapping of Unified Sentience”) as the sole inventor. The USPTO refused them, and in August 2022 the US Court of Appeals for the Federal Circuit agreed, holding in Thaler v. Vidal, 43 F.4th 1207 that the Patent Act’s plain text requires an inventor to be a natural person. The Supreme Court declined to hear the appeal in April 2023, leaving that ruling as the law.

The same challenge was run in parallel around the world, and it lost almost everywhere:

• United Kingdom — On 20 December 2023 the UK Supreme Court unanimously ruled in Thaler v Comptroller [2023] UKSC 49 that an inventor must be a natural person, so DABUS could not be named and no patent could be granted on that basis.
• European Patent Office — The EPO’s Legal Board of Appeal rejected the DABUS applications, holding that a designated inventor must have legal capacity, which an AI machine lacks.
• The outlier — South Africa granted a DABUS patent, but through a formalities-only system that does not substantively examine inventorship, so it is widely treated as an anomaly rather than a precedent.

Taken together, the DABUS litigation gives a rare thing in emerging tech law: a consistent, cross-border answer. Wherever a patent office substantively examines who invented something, the answer to can AI be an inventor has come back no.

The USPTO’s 2025 Reversal: AI as a Tool, Not an Inventor

US policy on the human side of the question moved twice in two years, and the current position matters. In February 2024 the USPTO issued detailed “Inventorship Guidance for AI-Assisted Inventions” that leaned on the multi-factor Pannu test to gauge whether a human’s contribution was significant enough.

That guidance did not last. On 26 November 2025 the USPTO published revised inventorship guidance that rescinded the 2024 version in its entirety, implementing Executive Order 14179 (“Removing Barriers to American Leadership in Artificial Intelligence”). The revised approach is deliberately simpler and more pro-filing:

• One standard for everything. “The same legal standard for determining inventorship applies to all inventions, regardless of whether AI systems were used” — there is no special AI inventorship test.
• AI is a tool. AI systems, including generative AI, “are tools used by human inventors”; using a tool does not elevate that assistance to inventor status.
• Presumption of human inventorship. The Office presumes the people named on the application are the actual human inventors, rather than starting from suspicion that AI did the inventing.

The bottom line did not change: only a natural person can be an inventor. What changed is the burden. The USPTO stopped asking applicants to prove, factor by factor, that a human did enough, and instead treats AI like any other instrument in the lab.

AI-Assisted Inventions: Who Actually Gets Named

Because can AI be an inventor resolves to no, the practical question becomes: which humans do you name when an AI system helped? Under the 2025 US approach the analysis is the ordinary one — the natural person or persons who conceived the claimed invention are the inventors, whether or not they used an AI tool to get there.

In practice, conception in an AI-assisted project usually lives in identifiable human acts: defining the problem and the constraints, designing or curating the training data, engineering the prompts or the objective, recognizing that a particular AI output was in fact the solution, and reducing it to a workable, claimed form. A person who does that is an inventor; a person who merely ran someone else’s model on someone else’s prompt generally is not.

Two mistakes recur. The first is naming no human at all because “the AI did it” — which, on the DABUS precedent, sinks the application. The second is over-naming, adding everyone who touched the project, which creates its own validity risk. The safe path is contemporaneous documentation: record what each human contributed to conception so the inventorship story survives later scrutiny in prosecution or litigation.

What This Means for Your AI Patent Filings

The inventorship rule is only the entry ticket. Two other forces decide whether an AI invention becomes a durable patent, and both show up the moment you map where to file. The first is subject-matter eligibility: a claim written as a pure algorithm or an abstract method risks a Section 101 rejection, while the same idea framed as a concrete system delivering a technical improvement is far more defensible.

The second is crowding. Generative AI is the fastest-growing field in the patent system — WIPO counted more than 54,000 GenAI inventions published between 2014 and 2023 — and the foundation-model core is densely owned. The room to file broad is usually one layer up, in deployment, efficiency and evaluation. That is exactly the pattern we mapped in our AI patent white space case study, where an applied-ML challenger sidestepped the owned core.

So the workflow for an AI invention runs: confirm a human inventor and document the conception, frame the claim as a concrete technical system to clear eligibility, and check the claim against the live filing record before you spend. Reading the AI patent landscape first is how a filing budget gets aimed at ground that is both open and defensible.

How the Inventorship Rule Shapes AI Patent Strategy

The stability of the answer is itself a strategic gift. Because every major office agrees a human must be named, companies can standardize one inventorship process worldwide instead of guessing at a patchwork — a rare certainty in a field where almost everything else is moving.

It also raises the value of good record-keeping. As AI tools do more of the heavy lifting, the defensible patents will be the ones whose files clearly show the human conception behind them. That evidentiary discipline is cheap to build now and expensive to reconstruct later, and it pairs naturally with the same rigor that separates ownership from a mere license — a distinction we cover in our guide to AI and copyright ownership.

The headline may keep asking whether machines can invent, but the operational reality is settled: humans invent, AI assists, and the patents that win are the ones filed by teams who understand both the rule and the crowded, eligibility-fenced landscape they are filing into.

How PerspireIP Can Help

At PerspireIP, our team helps innovators and businesses protect what they build. Whether you need a patent or trademark search, prior-art analysis, or an IP strategy tailored to your goals, we can help. Contact our team to discuss your next step.

Frequently Asked Questions