Can an Invention Made With AI Be Patented?

A New Question: Who Is the Inventor?

As generative AI takes on more of the research and development process, one question becomes harder to avoid: can an idea generated with AI's help be protected by a patent?

However strong the underlying technology is, failing to answer that question can undermine everything a company has built. Here, we walk through where the U.S. and Japan currently stand, based on primary legal sources.

The Shared Starting Point: Inventors Must Be Natural Persons

The first thing worth establishing is that both the U.S. and Japan agree on one point: an AI system itself cannot be named as the "inventor" on a patent.

This isn't a theoretical question. It was actually tested, and settled, when patent applications naming an autonomous AI system as the inventor were filed simultaneously in multiple countries, including the U.S., the U.K., Europe, South Korea, and Japan. None of them were accepted. The patent system is built around human creative activity, and treating an AI system itself as the rights-holding subject simply isn't contemplated under current law, in any of these jurisdictions.

That said, this doesn't mean an invention developed with AI can never be patented. The more accurate statement is: whether an AI-assisted invention can be patented depends on how substantially a natural person was involved, not on whether AI was involved at all.

Where the U.S. Stands

In 2022, the Court of Appeals for the Federal Circuit confirmed that under U.S. patent law, an "inventor" must be a natural person. An AI system itself cannot be listed as the inventor, but an invention that received AI assistance can still be patentable, as long as a natural person made a significant contribution.

Since then, the U.S. Patent and Trademark Office (USPTO) has issued practical guidance on how examination should handle AI-assisted inventions. One thing we discovered while preparing this article is worth mentioning directly: the guidance we initially referenced had already been superseded by a newer version. The most recent revision moves away from applying a special test just because AI was involved. Instead, it applies the traditional "conception" standard, whether a natural person formed the definite idea of the invention, uniformly, regardless of whether AI was used. AI is now treated the way lab equipment or a database would be: a tool that assists the inventor, not a co-inventor.

The specific test changed, but the underlying logic didn't. What matters isn't how AI was used, but what a human being conceived and decided. That thread runs through both the old and the new guidance.

Where Japan Stands

Japan's basic position matches the U.S. Japan's Patent Act has no explicit definition of "inventor," but the consistent administrative and judicial practice treats the inventor as necessarily a natural person.

The same international test case mentioned above was also filed in Japan, and the Japan Patent Office rejected it on formal grounds. That decision was upheld by the Tokyo District Court, and in January 2025, Japan's Intellectual Property High Court affirmed it as well, holding that AI-generated inventions cannot be granted patent protection under current Japanese law. The court also noted that whether AI-generated inventions should be patentable at all is a question better left to legislative debate, given its broad social implications. In other words, the door hasn't been permanently closed; it has been left open for future lawmaking.

The Practical Takeaway: What to Document

In both jurisdictions, what ultimately matters is what a human being actually did. Given that, it's worth keeping ongoing records, during the development process itself, of things like:

What specific technical problem a human defined
What a human specifically did with the AI (prompt design, constraints specified, how candidates were narrowed down)
What criteria a human used to evaluate and select among the AI's outputs
What a human ultimately added or changed

This isn't only relevant when a patent filing is on the horizon. Being able to explain, after the fact, what a human's role actually was in an AI-assisted process is itself a record that speaks to the quality of the development process as a whole.

Because This Field Moves Fast

One thing this article drove home while we were preparing it: rules around AI and intellectual property can be revised on a timescale of months, not years. Even expert commentary can be built on information that has quietly become outdated, if you don't check when it was current.

That's exactly why going back to primary sources, court decisions, official guidance documents, rather than relying on secondhand summaries, has to be part of how anyone works in this space.

How to Start a Conversation

"We want to keep using AI aggressively in development, but we're not sure how to protect our IP." That concern comes up for any company incorporating AI into R&D, regardless of size. At Mathematical Physics Labo, we're grappling with the same question ourselves, as a team that uses AI heavily in our own development work. The legal judgment calls around protecting IP belong to patent attorneys and lawyers, that's not something we offer. But when it comes to the operational side, how AI is used day to day, what gets recorded, and what gets kept out of AI tools altogether, that's something we might be able to talk through with you. If that's useful, we'd welcome the conversation.