When an AI design tool generates a circuit topology and an engineer later patents it, the question is not whether the circuit works. It is whether the engineer invented it. Under the law, "invented" means a human being formed a definite and permanent idea of a complete, operative invention. An AI generating the topology is not the same as a human conceiving it, no matter how much the human contributed.
The USPTO's revised inventorship guidance, issued November 28, 2025, makes this explicit in the Federal Register. Only natural persons may be listed as inventors. AI systems are tools that assist in the inventive process and do not qualify as inventors. The guidance rescinded and replaced the USPTO's previous approach from February 13, 2024.
This did not happen quietly. It happened in the Federal Register, which is not a publication most independent inventors read regularly.
The Two-Tier Problem
The guidance creates what patent attorneys are starting to call a two-tier innovation economy. Large firms with in-house counsel can document human conception at every step of an AI-assisted design process. They maintain records showing which engineer reviewed which AI output, when, and what they changed. They have the infrastructure to build the paper trail the USPTO now implicitly expects.
Solo inventors and startups do not have this infrastructure. When an independent engineer uses an AI tool to generate a design, iterates on it, and files a patent, the conception question becomes murky. The AI may have generated the operative idea. The human may have validated, refined, and implemented it. Under the new guidance, the question is whether the human conceived the invention under the traditional standard: formation of a definite and permanent idea of a complete, operative invention. But the guidance offers no framework for how to document that standard when AI generated the starting point.
"The juxtaposition is stark," wrote Matt Carey of Marshall Gerstein & Borun LLP: "The law is holding fast to human-only inventorship rules, even as the technology producing patent-worthy ideas is becoming increasingly autonomous."
The inventor community publication Inventors Spot noted that the November 2025 guidance "created a documentation requirement almost no one in the inventor community has noticed yet." That buried requirement may be the most consequential part of the entire document for the people most affected by it.
What the Guidance Actually Says
The core rule is straightforward: only natural persons qualify as inventors on US patent applications, including AI-assisted inventions. AI systems are tools only. No separate or modified inventorship standard applies to AI-assisted inventions.
The more granular rules are where the complexity lives. For single-person AI-assisted inventions, the inquiry is whether that person conceived the invention under the traditional conception standard. This sounds manageable until you ask what "conceived" means when an AI generated the topology and the human chose among variants.
For priority claims, the guidance adds another layer: claims must include at least one overlapping natural person as inventor. Claims based solely on AI inventorship will not be accepted. This matters for continuation applications and for patents that trace back to earlier filings where the human's role may have been even more attenuated.
The Pannu factors (the multi-factor test courts developed for determining whether a human contributor rose to the level of joint inventor) explicitly do not apply to single-person AI-assisted inventions. The guidance removes them from the analysis, which simplifies things for solo filers in one sense but also removes the framework some practitioners had been using to document human contribution.
The Precedent Behind the Policy
The legal foundation predates the guidance by three years. In August 2022, the US Court of Appeals for the Federal Circuit affirmed in Thaler v. Vidal (43 F.4th 1207) that only a natural person could be an inventor. Stephen Thaler's AI system DABUS had generated two inventions; Thaler listed DABUS as inventor on the patent applications. The court rejected the argument that the Constitution's inventorship requirements should accommodate AI inventorship. The ruling was not close.
The November 2025 guidance codifies the Federal Circuit's position and attempts to give practitioners a workable standard. Whether it succeeds is a different question.
What This Means for Builders
For engineers and founders building with AI-assisted design tools, the practical implication is documentation discipline that most independent practitioners do not currently have. The moment you use an AI to generate a circuit, a protein sequence, a mechanical structure, or a software algorithm, you are operating in a zone where your human contribution to conception needs to be demonstrable, not just present. It needs to be demonstrable in a way that survives a USPTO inquiry or an interference proceeding.
This does not mean AI-assisted invention is impossible to patent. It means the patentability question now includes a documentation burden that correlates with legal infrastructure. Large firms will build processes to meet it. Small filers will either develop similar processes or risk patents that do not survive scrutiny.
The irony is not lost on practitioners: the technology that makes independent invention more capable is also making the documentation of human invention more demanding. A solo engineer with an AI design tool can now generate ideas that previously required a research team. The patent system has not caught up to what that means for who can actually claim them.
What to watch next: how the USPTO handles the first significant challenge to a patent where the AI's role in conception is disputed. The guidance gives examiners a standard; it does not give them a test for how much human direction is enough.