USPTO AI Inventorship Guidance: What Changed in 2026

The USPTO AI inventorship guidance changed in November 2025, and if anyone on your team used an AI tool while developing an invention, it changes how you should be documenting that work right now. The old rule made AI-assisted inventions harder to qualify for a patent than ordinary ones. The new rule doesn’t. But it puts the burden back on you to prove, on paper, that the invention was yours before the AI ever entered the picture.

Where this started: only humans get to be inventors

The starting point hasn’t changed. A patent inventor has to be a natural person. That rule comes from Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), the case where Dr. Stephen Thaler tried to list his AI system, DABUS, as the sole inventor on a patent application. The Federal Circuit said no. Only a human being can be an inventor under the Patent Act, full stop. That holding never moved.

What’s moved twice since then is the rule for the much more common situation: a human invents something with help from an AI tool, and someone has to decide whether that human did enough of the inventing to qualify. If your team is using AI anywhere in R&D, our post on AI and intellectual property ownership covers the broader ownership picture this guidance sits inside.

The 2024 rule the USPTO just walked back

In February 2024, then-Director Kathi Vidal issued guidance for exactly that situation. It borrowed a legal test that was never built for it: the Pannu factors, a framework courts use to sort out joint inventorship disputes between two or more people. Under the 2024 guidance, if you used an AI tool while inventing something, you had to show you personally made a “significant contribution” to every claim, measured, in effect, as if the AI were a co-inventor whose share of the credit had to be argued down.

Patent attorneys flagged the problem with this from the start. An AI system isn’t a person, so applying a test built for person-versus-person disputes never fit cleanly. It also created uncertainty for exactly the kind of R&D the USPTO says it wants to encourage. AI-assisted invention is now routine in software, biotech, and materials science, and inventors had no reliable way to know in advance how much AI involvement would sink an application.

The new USPTO AI inventorship guidance: one test for every invention

On November 26, 2025, USPTO Director John Squires rescinded the 2024 guidance, and the USPTO AI inventorship guidance that replaced it runs to a single sentence of principle. The Federal Register notice, 2025-21457, published two days later, lays out the replacement, and it’s a simplification: the same legal standard applies to every application, whether or not AI was involved anywhere in the process. The Pannu factors go back to doing only the job they were built for, resolving disputes between multiple human co-inventors, because, as the guidance puts it plainly, an AI system is not a person and cannot be a joint inventor. The USPTO’s own AI policy hub collects the current guidance and the notices that preceded it, if you want more detail than this post covers.

In place of the old contribution test, the USPTO falls back on the traditional conception standard that’s applied to every patent application for decades: an inventor has to have formed the “definite and permanent idea of the complete and operative invention” in their own mind, before it gets reduced to practice. AI use doesn’t disqualify an applicant from meeting that standard. It also doesn’t excuse them from it.

The timing lines up with a broader policy push. The revised guidance implements Executive Order 14179, the administration’s directive aimed at removing regulatory friction around American AI development. This is one of several USPTO actions in the same vein, alongside a March 2026 bulletin examining generative AI’s effect on design-patent practice.

What “conception” means when an AI tool is involved

This is the part worth sitting with, because it’s where applications will succeed or fail. The guidance draws a line between two situations that can look similar from the outside.

A patent examiner, and later, if it’s contested, a court, is going to want to see that a human had the specific problem and a specific hypothesis for solving it before the AI tool produced anything useful, not that a human typed a vague prompt and kept whatever came back. The difference between “I asked an AI to design a better bracket” and “I identified that the bracket’s failure mode was fatigue at this specific stress point, and used an AI tool to help me model three candidate geometries that would relieve it” is the difference between an invention that’s yours and one that arguably isn’t.

Practically, that means the judgment calls you make after the AI produces something (which output you picked, what you changed, why you rejected the alternatives) are themselves part of your conception, and they’re the evidence that will matter if inventorship is ever questioned. An applicant who can’t point to any of that, and whose only real contribution was accepting an AI’s first answer, has a much weaker inventorship story under either the old rule or the new one.

What this means for your next application

If your team is already documenting AI-assisted development for other reasons, engineering notebooks, design history files, internal AI-use policies, you may already have most of what you need. If you aren’t, this is the moment to start, for three reasons that all point the same direction.

First, the guidance itself flags that “actual” inventorship and “legal” inventorship aren’t guaranteed to line up neatly, and IPWatchdog’s coverage of the rescission notes that some patent commentators expect courts to keep scrutinizing that gap even though the USPTO has simplified its own internal test. Second, claims that map cleanly to a documented human conception are easier to defend later, whether in prosecution, in litigation, or in an inventorship challenge from a competitor. Third, if more than one person touched the AI’s output, one person ran the tool, another refined the result, that’s now squarely a human-to-human Pannu situation, and it needs to be evaluated as one.

None of this changes what makes an invention patentable in the first place. The USPTO AI inventorship guidance changes what you need to be able to prove, and when you need to start proving it.

FAQ

Can I list an AI system as an inventor on my patent application?

No. Thaler v. Vidal settled this in 2022, and nothing in the 2025 guidance revisits it. Only a natural person can be a named inventor.

Do I have to disclose that I used AI when I file?

The USPTO doesn’t require a blanket AI-use disclosure, but your duty of candor to the office still applies to any facts that bear on inventorship. Talk to your attorney about what that means for your specific application.

Does this affect patents I already filed under the 2024 guidance?

It’s worth a second look, especially for any application where inventorship was framed around a “significant contribution” analysis. The new guidance doesn’t automatically invalidate anything already filed, but it changes the standard examiners and courts will apply going forward.

Does this make it easier to patent something I built with AI help?

Not automatically. It removes an AI-specific hurdle that didn’t fit the situation well, but you still have to clear the same conception standard every other applicant clears. What’s different is the kind of evidence that will settle the question if it’s ever raised.

Knowing what the USPTO AI inventorship guidance says is one thing. Building a paper trail that actually holds up if an examiner or a court ever questions your inventorship is a different skill, and it’s the kind of thing worth setting up correctly from your very next invention disclosure rather than reconstructing after the fact. A patent attorney who’s watched how examiners are applying this guidance in practice can tell you exactly what to start documenting today.

Get Your Free Patent Needs Assessment