On March 4, 2026, the Supreme Court dismissed Stephen Thaler's appeal in the DABUS AI case and let stand a 2025 IP High Court ruling that an artificial intelligence system cannot be named as the inventor on a patent application, but the decision did not address who can claim…
The Supreme Court of Japan closed the country's last legal door on naming an artificial intelligence as a patent inventor on March 4, 2026, when it dismissed Stephen Thaler's final appeal without issuing a substantive opinion of its own. The decision lets stand a January 30, 2025 ruling from the IP High Court, Japan's specialized patent appeals court, which held that under the Patent Act an inventor must be a natural person. With Japan's highest court declining to disturb that holding, the DABUS challenge ends the same way it has ended in six other jurisdictions: the AI itself cannot be credited on paper as the author of an invention.
The case turned on two Japan national-phase patent applications that Thaler, a United States national, filed in August 2020. Both listed a machine called DABUS (Device for the Autonomous Bootstrapping of Unified Sentience) as the sole inventor. The applications originated from PCT/IB2019/057809, filed September 17, 2019, and claimed subject matter including a food container and related devices and methods designed to attract attention. The Japan Patent Office rejected the inventor designation in a notice dated July 30, 2021, formally dismissed the application on October 13, 2021, and turned down an administrative review in October 2022. The Tokyo District Court ruled against Thaler on May 16, 2024, and the IP High Court affirmed under presiding judge Hibiki Shimizu on January 30, 2025, before the Supreme Court's final dismissal.
The substantive reasoning that doomed Thaler's case crystallized at the IP High Court level, not the Supreme Court, because Japan's top court accepted the lower court's logic without restating it. According to analyses from Japanese patent practitioners including Keisen Associates and Yuasa & Hara, the Patent Act operates on two simultaneous tracks. Substantively, it grants rights only to inventions attributed to a natural person. Procedurally, it treats the applicant's Shimei (the name of the natural person who is the inventor) differently from Meisho, the entity name that may stand for an applicant or rights holder. An AI system, lacking legal personality, cannot fill the Shimei slot. The court also read Articles 29(1) and 35(3) of the Patent Act as a closed attribution chain that runs only from natural persons to invention, leaving no path by which a machine acting alone could originate a patentable idea.
Japan joins a long line of jurisdictions that have refused the same request. Thaler has pursued DABUS inventorship in the United States, the United Kingdom, the European Patent Office's member states, Australia, South Africa, and China, and in every case the AI-inventor designation has been rejected. The convergence matters less as a vote count than as a structural observation. Every patent system the case has reached defines the inventor as a natural person at the point of original attribution, and none of them has been persuaded to bend that definition. Yomiuri's Japanese-language coverage and the Asahi Shimbun both report the Supreme Court dismissal in the same terms, and the Mainichi Shimbun's English account of the IP High Court ruling describes an analogous framework. The pattern is consistent: the world is not granting AI patents.
The sharpest objection to the ruling, surfacing in online discussion including a Hacker News thread about the case, is this: the systems the courts say cannot be inventors are trained on human-authored text, code, and imagery whose provenance is rarely disclosed or compensated. The machines have human authors, but what they produce has, legally, no author. That tension does not undo the ruling, because patent law is a different framework from copyright and trade-secret law. It does mean the legal verdict and the commercial reality are now further apart than they were a decade ago, and the gap will keep widening as AI moves deeper into research and development.
The harder question the Supreme Court left unresolved is who legitimately owns an AI-assisted invention. The IP High Court's structural answer implies a default: if DABUS cannot be the inventor, and Thaler denies being one, the natural person who operated the AI becomes the inventor by necessity, not by legal design. Whether the operator can claim invention under Article 29(1), which conditions patentability on the contribution by a natural person, depends on facts Japan's ruling does not address: what the operator prompted, what the AI generated, and what the operator selected or revised. Yomiuri's coverage and the Legal500 analysis both emphasize that any change to admit AI-generated inventions into the patent system would have to come from the Diet, the Japanese legislature, not the courts.
What to watch next is whether Japan's legislature moves first or waits for a major foreign patent office to amend its rules. The United States is the likeliest bellwether: if the U.S. Patent and Trademark Office ever loosens inventorship requirements for AI-assisted inventions, the Diet's calculus changes overnight. Until then, the working assumption for anyone using AI in research and development (and Thaler, his critics, and the Japanese patent bar all agree on this much) is that a human must sign. That is the line Japan has now drawn, and it is the line every patent counsel is now operating inside.