On November 28, 2025, the United States Patent and Trademark Office quietly rewrote the rules governing what it means to invent something with the help of AI. The change reversed a legal standard that had been building exposure for independent inventors since early 2024, a standard most of them never knew existed. It also introduced a new vulnerability that the inventor community hasn’t absorbed yet.
Most inventors using AI tools have been operating under a reasonable assumption. They’re the human. They’re the inventor. The AI is a tool.
That assumption is now legally correct. It wasn’t always.
The Rule That Created A Hidden Trap
In February 2024, the USPTO issued guidance on AI-assisted inventions. The agency’s position was clear: only humans can hold patents. But in implementing it, the USPTO made a move that IP attorneys quietly flagged as problematic.
It applied the Pannu factors to AI-assisted inventions.
The Pannu factors, drawn from Pannu v. Iolab Corp., are a legal test designed for joint inventorship between multiple humans. When two people work on a patent together, each must demonstrate a significant contribution to at least one claim. The 2024 guidance borrowed this test and applied it to solo inventors working with AI, requiring that a single person “make a significant contribution to every claim” even when the only other participant was a software system.
The practical effect: the more meaningfully an inventor used AI, not to search prior art or clean up language but to actually explore the invention space, the more vulnerable their patent became to a challenge arguing insufficient human contribution.
Most independent inventors never knew this was the standard. They were filing patents, using AI throughout their process, and sitting on a quiet exposure.
What The November Guidance Actually Says
The Federal Register notice published under new USPTO Director John Squires is unambiguous: “The Pannu factors only apply when determining whether multiple natural persons qualify as joint inventors. Pannu is inapplicable when only one natural person is involved in developing an invention with AI assistance, because AI systems are not persons and therefore cannot be joint inventors.”
AI is a tool. The same as a spectrometer, a lathe, or a CAD program. The inventorship standard is the same whether AI was involved or not.
As IPWatchdog noted at the time, the new guidance explicitly rejects the approach taken under the previous administration, an unusual public break from a prior position. The Pannu-factors framework is gone.
For inventors who had been using AI tools throughout their development process, that’s real relief.
The Exposure The New Rules Create
What the new guidance puts in place is different from what it removes, and it’s where independent inventors need to pay attention.
With the Pannu test gone, inventorship analysis returns to its traditional foundation: conception. Courts have long defined it as “the formation, in the mind of the inventor, of a definite and permanent idea of the complete and operative invention.” The question is now simply whether a human formed that idea. Not whether the human contributed enough relative to the AI. Whether the human conceived it at all.
That sounds like a lower bar. It is, mostly. But conception is harder to prove when AI is in the room.
When an inventor works through a problem with generative AI, the process leaves a record that can look, on its face, like the AI generated the invention and the human approved it. Prompt. Response. “Looks good.” That’s a thin paper trail for a conception claim. IP attorneys reviewing the December guidance consistently flagged the same issue: in an inventorship dispute, the question will be whether a human formed the definite idea, and if the only record is an AI conversation log showing the model proposed a concept and the inventor agreed, that’s a problem.
The history of what inventors do versus what scientists and engineers do is relevant here. The inventor’s act is one of conception and recognition. That act can happen with AI assistance. But it has to be traceable to a human mind.
What protects inventors isn’t the absence of AI in the process. It’s evidence that the human was the one thinking: framing the problem, making decisions, choosing directions, setting constraints. The prompts an inventor writes to an AI system, the choices they make when the model returns options, the reasoning behind what they keep and what they discard, all of that is conception. Dated notes, exported conversation threads, and brief records of decision reasoning are what make that visible to a patent examiner or a court.
Inventors who treat the new guidance as a green light to use AI without thinking about documentation are misreading it. The old rules created one kind of exposure. The new rules create a different one, quieter and probably more dangerous because it’s easier to miss.
That dynamic tracks the longer history of how inventors have always had to prove their contribution in ways scientists don’t. The documentation discipline hasn’t changed. What’s changed is what needs to be documented.
My Take
The November guidance is the right call. Applying a joint-inventor test to a solo inventor working with a software tool was a category error, and the USPTO corrected it before it caused widespread damage. But the inventor community’s general response, to the extent it has noticed at all, has been to treat this as permission to stop worrying. That’s the wrong read. The inventors who benefit most from the new rules will be the ones who understood the old rules well enough to recognize what actually changed, and who start keeping records of their conception process before they need to prove it in a dispute.
