When AI Is the Inventor, the Patent Problem Starts Before Filing

Imagine a lab platform that combines a generative model, robotic automation and an active-learning loop. The platform proposes experiments, runs them, reads the results and iterates on its own. Eventually, it finds a novel enzyme mutation that boosts activity twenty-fold. No scientist proposed the mutation. No scientist identified the specific sequence beforehand. A human being pressed “start,” then reviewed the answer at the end. Under the current U.S. patent law, that is not just a difficult inventorship case. It may be a no-inventor case. That means the problem is not how to list the inventors more carefully. The problem is that there may be no valid human inventor to list at all, which will result in no patent.

That conclusion follows from the basic rule the U.S. Patent and Trademark Office (USPTO) and the Federal Circuit have now made unmistakably clear: only a natural person can be an inventor. In Thaler v. Vidal (2022), the Federal Circuit held that the term “inventor” refers to a human being. More recently, the USPTO’s November 2025 revised inventorship guidance reaffirmed that principle and clarified that the same legal standard applies regardless of whether AI is involved—the touchstone is still human conception. AI systems are treated as tools, not inventors.

So what happens in the hypothetical above? If no human conceived the claimed mutation, then there is no lawful inventor. In that circumstance, a patent cannot be validly obtained for that subject matter. And trying to “fix” the problem by listing a human who did not actually conceive the invention does not cure the defect. It creates a new one. The resulting patent is vulnerable to invalidity challenges, and potentially even unenforceability arguments if inventorship was knowingly misrepresented.

This is why the real risk in AI-assisted invention is not the use of AI itself. The risk is that you cannot prove what the human actually invented. If the human contribution cannot be tied to the conception of the claimed subject matter, the patent system has nothing to reward.

The solution is not to limit AI use. Nor is it realistic. It is to design the innovation process so that human conception is real, identifiable, and documented to prove it. One effective approach proposed here is to adopt a structured three-step documentation framework: a Conception report, a Work report, and an Invention Disclosure Form.

Conception Report: Defining the Human Contribution Upfront

Before using AI, scientists should prepare a short report that captures the human framework for the AI use. This report should address:

• What specific product or invention are you trying to develop, what known product or invention is the closest point of comparison, and what quality or function do you expect your product or invention to have in comparison?
• What specific scientific or technical problem do you want the AI to help you solve?
• What limits, requirements, or design criteria have you decided to set for the AI? Why do you think they will help reach your goal?

The purpose is to document human conception before any AI is used to develop the invention—not after the fact—and to document it in a way that supports a human inventorship narrative. By defining the problem, constraints, approach, and outcome, the scientist establishes a baseline of conception for inventorship.

Work Report: Documenting the Process

The Work Report captures the AI-assisted work as it unfolds—the prompts, responses, iterations, and judgment that shaped the result. How much of that to preserve, though, cuts in two directions. On the one hand, logging everything, like an electronic lab notebook, builds a complete record that can corroborate the scientist’s conception. On the other hand, inventorship challenges are hard for an opposing party to win given the high burden of clear and convincing evidence in district court. A granular prompt log can potentially be recast against the inventor. Therefore, a company should carefully consider how much and the types of data that should be kept in such a log, while recognizing that just having a documented “plan” with no prompts may carry its own optics problems. The better solution is to have a blanket policy that keeps some data, but perhaps not all. For example, the number of results that were returned or iterations required to generate the final result, and documentation regarding how the final results were chosen by the human.

To make this framework operational, companies may consider adopting a simple but effective internal policy: a scientist may not request a patent filing unless the scientist has completed a defined number of AI iterations—whether five, 10, or more.

The point is not to impose arbitrary hurdles but to require enough engagement to explore the problem space and identify the strongest solution—and to show that the inventor shaped the result by refining inputs, evaluating outputs, and exercising judgment rather than merely accepting an AI-generated answer.

Invention Disclosure Form

When scientists determine that patent protection may be needed for a research outcome, they should submit an invention disclosure form (IDF). The IDF should require attachment of the Conception Report and Work Report, along with an explanation of how many AI iterations were performed and why. When those reports are prepared in a structured way, much of the IDF can be completed directly from the reports.

In addition to the reports, the IDF should include focused questions that ask scientists to describe, in their own words, the decisions they made and the scientific judgment they exercised during the AI-assisted process. Examples include:

• Which AI-generated result are you seeking to protect, and which other results did you decide not to pursue? Please explain why.
• What tests or validation steps did you perform, and why did you choose those steps instead of others?
• What additional experiments, tests, or validation steps do you plan to perform in the next 12 months, and what do you expect they will show beyond the data already generated?

These questions are designed to capture human contribution as part of the ordinary course of invention development. That documentation process helps preserve a clear record of the scientist’s role, supports patent filings that properly identify human inventors, and reinforces that the scientist is directing the process rather than merely observing an AI-generated result.

Making the Process Stick: Training

To make this framework effective, companies should implement regular training for scientists and other innovation personnel. That training should be updated as AI technology evolves so that scientists understand how new tools affect the inventorship analysis and the documentation required to preserve patent rights. Repeated training reinforces why these steps matter, helps scientists identify the kinds of human contributions that must be documented, and makes the process a routine part of AI-assisted research rather than an after-the-fact exercise.

Conclusion

AI does not change the rule any more than a microscope does: patent rights still depend on human planning and conception. The answer is better process—define the human contribution upfront, document the work that supports it, train scientists to recognize what matters and capture it before filing. Companies that adapt will not only preserve patent exclusivity—they will move ahead of competitors that fail to build human inventorship into the process.

Reprinted with permission from the June 18, 2026, issue of Corporate Counsel. © 2026 ALM Media Properties, LLC. Further duplication without permission is prohibited. All rights reserved.