Can an AI Be Named as an Inventor? Where Every Major Patent Jurisdiction Stands in 2026

On 4 March 2026, Japan’s Supreme Court dismissed the final appeal in the DABUS litigation, ending a legal campaign that spanned seven years and 18 jurisdictions. The question it settled sounds academic: can an artificial intelligence system be named as the inventor on a patent application? The answer, delivered with remarkable consistency across every court of final instance that has considered it, is no.

But the settled question is not the interesting one. If your R&D team uses generative AI, and in 2026 almost every R&D team does, the question that determines whether your patents survive is not whether the AI can be the inventor. It is what the humans around the AI must have done, and that question now has materially different answers in Washington, Munich, and Tokyo.

This is the inventorship companion, or Part 2, to Can You Patent Software, an Algorithm, or an AI Model?. That guide asks whether AI and software subject matter can qualify for patent protection. This article asks who can legally claim to have invented it.

The DABUS test case in one paragraph

In 2018, Dr Stephen Thaler filed patent applications for two inventions, a fractal beverage container and a pulsing light beacon, naming his AI system DABUS as the sole inventor. He filed deliberately and transparently, arguing that he owned the AI’s output the way a farmer owns fruit from his tree, and that naming himself as inventor for work he did not perform would be dishonest. From a single PCT application, the case spread across the world. It became the closest thing patent law has ever had to a controlled global experiment, because every court answered the same facts under its own statute. A WIPO study of inventorship issues describes the international application and its filings in 17 national and regional jurisdictions.

Jurisdictions with final, unappealable decisions: AI cannot be the inventor

Ten patent systems have now closed the question with no further appeal available.

Jurisdiction	Final decision	Year closed
United States	Thaler v Vidal, Federal Circuit; certiorari denied	2023
United Kingdom	UK Supreme Court, [2023] UKSC 49	2023
European Patent Office	Legal Board of Appeal, J 8/20; divisional refusal upheld	2021, 2026
Australia	Full Federal Court; special leave to appeal refused	2022
Germany	Federal Court of Justice, X ZB 5/22	2024
Taiwan	Supreme Administrative Court	2023
Saudi Arabia	Formalities acceptance reversed on review	2024
Israel	Tel Aviv Yafo District Court	2025
Japan	Supreme Court dismissal of final appeal	2026
Switzerland	Federal Administrative Court (partial exception, discussed below)	2025

The reasoning converges on the same statutory point everywhere: patent acts were written with a natural person in mind. The US Federal Circuit held that an inventor must be an individual, meaning a human being. The UK Supreme Court reached the same conclusion under the Patents Act 1977, and added that Dr Thaler could not claim ownership of the inventions through the doctrine of accession, because accession does not apply to intangible property. Japan’s IP High Court, whose reasoning the Supreme Court left in force, went one step further and stated explicitly that recognising AI inventorship would require new legislation, not judicial reinterpretation. An English translation of the Japanese judgment is available from the court.

That last point deserves emphasis for anyone tracking this area. No court anywhere concluded that AI inventorship is impossible in principle. Every court concluded that it is a question for legislatures. The consensus is statutory, not philosophical, which means it can change, but only through parliaments.

The one genuine outlier: South Africa

South Africa granted patent ZA2021/03242 in July 2021 with DABUS listed as the inventor. It remains the only granted patent in the world naming an AI system as inventor after substantive review elsewhere failed everywhere.

The asterisk is significant. South Africa operates a depository system. The Companies and Intellectual Property Commission examines formalities only and conducts no substantive examination of novelty, inventive step, or, as it turned out, inventorship. No South African tribunal has ever held that an AI may be an inventor, and local practitioners have argued publicly that the grant was defective even on formalities grounds. Revocation grounds under the Patents Act 1978 remain available. The grant carries essentially no persuasive weight internationally and should not inform anyone’s filing strategy.

Still open: the appeals pipeline

Five jurisdictions have rejected the applications but appeals remain pending: Canada, following the Commissioner’s refusal of June 2025; China, before the Beijing IP Court; South Korea, where the refusal was upheld by the Seoul Administrative Court in 2023 and the Seoul High Court in May 2024, with a further appeal outstanding; New Zealand, where leave to appeal a 2023 High Court refusal was granted; and Brazil.

India’s Patent Office issued its first substantive determination on 15 April 2026, in two separate orders: a refusal under Section 15 of the Patents Act 1970, and a disposal of a pre-grant opposition under Section 25(1). Notably, the Indian refusal rested on two independent grounds, non-compliance with the inventorship provisions and lack of inventive step, making it one of the few DABUS decisions to also reject the invention on its merits. It is an administrative decision rather than a judicial one, so appeal routes remain open. Singapore’s application is still pending examination.

Nobody following these cases expects a different outcome in any of them. The value of the pipeline is that each decision adds another data point on the reasoning, and the reasoning is where jurisdictions have started to diverge.

The divergence that matters: what must the human have done?

Here is where the story stops being uniform, and where filing strategy gets genuinely difficult for AI-assisted inventions.

The United States applies a pure conception standard, and it changed recently. The USPTO first addressed this in February 2024, adapting the Pannu factors, a joint-inventorship test, to ask whether a human made a significant contribution to an AI-assisted invention. That guidance was rescinded in its entirety and replaced by the Revised Inventorship Guidance for AI-Assisted Inventions, published in the Federal Register on 28 November 2025 and effective immediately.

The revised framework is deliberately simpler. It reaffirms that only natural persons can be inventors. It treats AI systems as instruments, no different in principle from laboratory equipment, software, or a research database: they may assist in creating an invention but cannot conceive one. It applies the traditional, fact-intensive conception test uniformly, with no special rules for AI assistance. And it drops Pannu analysis where a single natural person is involved, retaining that test only for joint inventorship among multiple humans, even where those humans used AI.

Anyone working from commentary written between February 2024 and late 2025 is working from a rescinded framework. The practical bar has arguably become easier to clear, since AI use no longer triggers a distinct analysis, but the underlying requirement is unchanged: a human must have conceived the invention.

Germany applies a causation standard. The Federal Court of Justice held in 2024 that while the named inventor must be human, that human need only have caused the AI to generate the invention. Human influence, the court said, need not itself be inventive or substantial. The person who configured and ran the system can be validly designated. A Swiss federal court followed similar reasoning in 2025, and the Swiss patent office did not appeal.

Japan sits closer to the American position. The inventor must satisfy the traditional conception requirement, forming a definite idea of the complete and operative invention.

Consider what this means in practice. The same invention, produced by the same workflow, can have a valid inventor designation in Munich and a defective one in Alexandria, Virginia. A researcher who prompted an AI system, curated its outputs, and selected the winning candidate may clear the German bar comfortably while failing the American one, depending on how much inventive judgment the selection involved. For a portfolio filed across both jurisdictions from a single priority application, the inventor declarations may need to be reasoned differently for each office, and the internal records supporting them need to exist before filing, not after.

And the German cautionary tale cuts the other way. In the fractal-container proceedings, an applicant named himself as inventor while also stating in the application that the invention was conceived by DABUS. The court treated the contradiction as invalidating the inventor designation. Transparency about AI involvement is fine. Contradicting your own inventorship declaration is not. The EPO reached the same result in its February 2026 divisional decision.

A final wrinkle: not every system forces the question. Under Section 39 of Israel’s Patents Law 5727-1967, naming the inventor is a right the inventor may demand rather than a mandatory application requirement. Israel still refused the DABUS application, but only because it affirmatively named an AI; an applicant who stays silent on inventorship never triggers the issue there.

What this means for anyone filing in 2026

Three practical conclusions follow from the case law.

First, AI-assisted inventions remain fully patentable everywhere. No jurisdiction has held that using AI in the inventive process disqualifies an invention. The consistent rule is that a natural person must be named, and that person must satisfy the local standard for inventorship. AI-related filing volumes have grown throughout the DABUS years, which tells you the doctrine was never the obstacle.

Second, contribution records are now a filing asset. The US conception test is fact-intensive, and inventorship defects can render a patent unenforceable. Teams using AI in R&D should document who directed the AI, who made the key selections and judgments, and when. This is the same discipline laboratory notebooks served in the interference era, revived for a new reason.

Third, watch legislatures, not courts. Every final court decision pointed at the same door: if AI inventorship is to exist, parliaments must create it. Japanese policymakers have already opened discussions on how the Patent Act should treat AI-assisted invention and the contributions required for human inventorship, as Japan’s Intellectual Property Strategic Program 2026 materials show. The judicial phase of this question is over. The legislative phase has barely begun.

The operational lesson is simple: treat AI as part of the invention record, not as the inventor. Preserve the human decisions that turned a possibility into the claimed technical solution, and test that record against every jurisdiction in the filing plan.

If your team is developing an AI-assisted invention, search the worldwide prior art in Patenta before drafting, then carry the closest references and the human contribution record into the application.