No. Under the AI inventor patent rules in the United States, only a human being can be named as an inventor, so an AI system cannot appear on a patent application or an issued patent. That has been the law since the Federal Circuit decided Thaler v. Vidal in August 2022, and the Supreme Court declined to review it in April 2023.

What can be patented is an invention a person made with AI help. In November 2025 the Patent and Trademark Office changed how it decides whether a person did enough, and the change is easier on inventors than the guidance it replaced. This post walks through the court ruling, the two sets of agency guidance, and how to think about listing inventors when AI was involved. It covers US patent law only.

The quick version

Question Answer in the US
Can an AI system be listed as an inventor? No
Can I patent something I invented using AI? Yes, if a natural person conceived the claimed invention
Does the AI’s owner count as the inventor? Not merely by owning or running the system
Does the USPTO apply a special test to AI-assisted inventions? Not since November 2025; the same test applies to all inventions
Has Congress changed the statute? I found no change

What Thaler v. Vidal decided

Stephen Thaler filed patent applications naming an AI system, which he called DABUS, as the inventor. The Patent and Trademark Office refused to accept that, a district court agreed, and the Federal Circuit affirmed in a decision dated August 5, 2022, written by Judge Stark for a panel that included Chief Judge Moore and Judge Taranto.

The reasoning was about the statute’s words. The Patent Act defines an inventor as an “individual,” and the court read that word to mean a human being. It therefore held that an AI system cannot be listed as an inventor. Thaler asked the Supreme Court to take the case, and it denied the petition on April 24, 2023.

The court was careful about the edges. It did not decide whether an AI system can form beliefs, and it noted that it was not presented with the question whether inventions made by human beings with the assistance of AI are eligible for patent protection. It also called Thaler’s policy arguments for patenting machine-made inventions speculative and unsupported by the text or record. Those arguments, in other words, are for Congress.

February 2024: the first USPTO guidance

On February 13, 2024, the agency published Inventorship Guidance for AI-Assisted Inventions at 89 Federal Register 10043. It confirmed that AI systems cannot be inventors and that AI-assisted inventions are not categorically unpatentable. It then told examiners and applicants to test each claim for a natural person who made a significant contribution, using a three-part framework from a Federal Circuit case called Pannu v. Iolab:

  1. The person contributed in a significant manner to conceiving the invention or reducing it to practice.
  2. The contribution was not insignificant compared with the whole invention.
  3. The person did more than explain well-known concepts or the current state of the art.

The guidance also offered principles. Using AI does not cancel a person’s contribution. Merely presenting a problem to an AI system was generally not enough, though how a person built the prompt might count. Reducing an invention to practice alone was not sufficient, and owning or supervising the AI system did not make someone an inventor. Applications naming an AI as inventor faced rejection, an AI could not sign the oath, and a foreign priority claim naming an AI as sole inventor would not be accepted.

November 2025: the revised guidance

On November 28, 2025, the agency published Revised Inventorship Guidance for AI-Assisted Inventions in the Federal Register. It rescinds the February 2024 guidance in full.

The new notice says the same inventorship standard applies to all inventions, whether or not AI was used. There is no separate or modified test for AI-assisted work. The central question is conception, which the Federal Circuit treats as a definite, permanent idea of the complete, operative invention: a specific, settled solution rather than a general goal. The test is whether the idea was clear enough in the person’s mind that ordinary skill alone would be needed to reduce it to practice.

The Pannu factors survive but in a narrower role. They apply only when several natural persons may be joint inventors, to decide whether each qualifies. AI use does not change the analysis among the human contributors.

The notice treats AI as a tool, comparable to laboratory equipment, software or a research database. An AI can generate ideas or provide services, but it stays a tool. Where one natural person is involved, the question becomes whether that person conceived the claimed invention. An application that names an AI system as an inventor should be rejected under 35 U.S.C. 101 and 115 for all claims or otherwise handled appropriately. The names given in the application data sheet or oath are presumed to be the actual inventors.

Several things did not change. For design patents the same inventorship approach applies. For plant patents, the inventor must have contributed to creating the plant through breeding or other agricultural or horticultural efforts, in addition to recognizing its uniqueness and reproducing it asexually, and that applies to AI-assisted varieties too. A priority claim must share at least one natural-person inventor with the earlier filing, and a foreign application naming only an AI as inventor will not be accepted. If a foreign application lists both a person and an AI, the US data sheet should list only the person or persons.

Two points the notice leaves open: it states no effective date in the text I read, and it does not discuss how prompting or other human input should be treated. It also gives no worked examples. Practitioners and applicants are left to apply ordinary conception law.

What this means for listing inventors

These steps follow from the sources above. They are general orientation, not a substitute for a patent attorney’s review of the facts.

  1. Name people, never the tool. Any application that lists an AI system as inventor is headed for rejection.
  2. Ask who conceived each claim. For every claim, identify the person who had the definite idea of the complete invention. A general goal handed to a system, with the system choosing the solution, is the weak case.
  3. Treat the AI like equipment. If a model suggested candidate designs and a researcher chose, understood and refined one into a specific solution, the researcher’s role looks like conception. If the model supplied the whole operative idea and the person only recognized it afterward, the position is less certain, and the new notice offers no example to settle it.
  4. Do not add people for appearances. Owning or overseeing the AI is not enough. Naming someone who did not conceive a claim can create problems later.
  5. Keep lab notebooks and prompt histories. Records showing who decided what are the best evidence of conception.
  6. Use the right test for teams. When several people worked on the invention, the Pannu factors still sort the joint inventors.

How the single-inventor question looks in three scenarios

Take an engineer who asks a model for ten possible heat-sink geometries, studies the results, and then specifies one particular geometry with the dimensions, materials and tolerances that make it work. The engineer arrived at a specific, settled solution, so conception rests with that person. Now take someone who only types a goal such as “make a better battery” and files whatever comes back. Under the notice’s focus on a definite idea of the complete invention, that person has the general goal and little else. The middle cases, where a person recognizes a good output and then refines it, depend on facts, and the agency has published no examples to draw the line. A patent attorney will ask what you decided, in what order, and what the records show.

What is still unsettled

The statute itself has not changed. A court could eventually hear a case about an invention where a person’s role was thin, and the Federal Circuit’s remark that it had not been presented with AI-assisted inventions leaves that door open. The agency’s guidance has also changed once already in under two years, which is a reason to check the current version before relying on any summary.

Other countries have heard the same DABUS applications. The United Kingdom Supreme Court handed down its judgment in Thaler’s case on December 20, 2023 ([2023] UKSC 49), after the Comptroller had refused to treat DABUS as a “person” who could be named. This post does not summarize that decision or other countries’ rules, so check those separately if you file abroad.

Related questions

Can an AI be an author, as well as an inventor? The US answer is parallel. The Copyright Act has been read to require a human author, as a separate post on AI-generated art explains.

Do trademarks work the same way? I did not research trademark ownership for this post. Trademark rights arise from use of a mark in commerce rather than creative authorship, so the inventor question does not map directly; ask a trademark attorney about a specific mark.

Does using ChatGPT while inventing taint my patent? Under the November 2025 notice, using AI as a tool does not by itself defeat inventorship. What counts is whether a natural person conceived the claimed invention.

This is general information, not legal advice. Patent filings carry deadlines and consequences, so talk to a registered patent attorney or agent.

Sources