AI-generated inventions raise questions that reach beyond intellectual property law. For industrial automation companies using artificial intelligence in research and development, the answers could affect how teams document technical work, manage patent rights and evaluate whether an AI-assisted result is genuinely inventive. The central challenge is distinguishing an invention that uses AI as a tool from one that an AI system functionally conceives on its own.
A 2019 submission to the U.S. Patent and Trademark Office (USPTO) proposed treating an AI-generated invention as one that would qualify as patentable and would have been conceived by a human inventor if the AI were a natural person. That definition is a policy proposal, not a settled rule. It nevertheless offers a useful framework for understanding the issues facing companies integrating AI into engineering and product development.
AI-assisted work is not necessarily AI-generated invention
AI has many roles in technical work. A system might analyze data, search technical literature, identify patterns or help a researcher compare design options. Those activities can support an invention without making the system its inventor. The more specific question is whether the AI contributed to the conception of the particular invention—the point at which a definite and complete idea of the invention has been formed.
That distinction matters in industrial settings. An engineer who uses a model to explore alternatives may make the inventive contribution through the problem definition, system configuration, selection of inputs or interpretation of results. In another case, a person may simply submit a problem and receive a solution that the system generated independently. The mere presence of AI in the workflow does not, by itself, resolve who conceived the invention.
Why assigning AI inventorship is difficult
Potential human contributors include the people who define the problem, develop or train the AI, prepare its data, operate the system, interpret its output or recognize a result as inventive. But involvement in the process is not automatically equivalent to inventorship. The submission argues that programmers should not be named as inventors solely because they built a system capable of solving problems, particularly if they did not conceive of the specific output. Likewise, simply asking an AI to solve a problem may not amount to conceiving the resulting solution.
Human judgment can still be decisive. A researcher may select a particular result from many outputs or apply technical skill to understand why an output could work. In those circumstances, that contribution may be relevant to inventorship. Conversely, if the inventive character of an output is immediately apparent, identifying it may not require the kind of contribution that qualifies a person as an inventor. These distinctions can be difficult to make, especially when several people and tools contribute across a distributed development process.
Patent ownership and the incentive question
If an AI system were recognized as an inventor, it could not itself own a patent under the approach described in the submission: AI lacks legal personhood and cannot own property. The author therefore proposed that patent rights should vest in the AI’s owner, subject to agreements that might allocate rights differently. Other possible candidates include the system’s developer or user, while open-source or distributed systems can make ownership less straightforward.
The argument for allowing patents on AI-generated inventions is primarily about incentives and disclosure. If organizations can protect qualifying inventions, they may have stronger reasons to invest in systems capable of generating them and to disclose the resulting technical knowledge through patent applications. The countervailing concern is that granting patents too readily could restrict competition or reward outputs that would arise routinely from standard research methods.
For manufacturers, automation providers and system integrators, the practical takeaway is to make ownership and contribution questions part of AI project governance. Agreements with developers, vendors, research partners and employees can clarify rights to outputs and related inventions. Teams can also keep records of the problem being addressed, the system’s role, human decisions and how a proposed solution was selected or developed. Such documentation does not determine inventorship on its own, but it can help explain the development process.
AI could also change the patentability threshold
Patentability is not only about who invented something. Under the nonobviousness test, an invention may be unpatentable if it would have been obvious to a hypothetical person with ordinary skill in the relevant field, considering the prior art. The submission argues that this benchmark should evolve as AI becomes a routine part of research. If researchers commonly use AI to analyze information and develop solutions, the capabilities of an AI-augmented worker may become relevant to what counts as ordinary skill.
The proposal considers a further possibility: where inventive AI becomes the standard research method for a field or problem, the benchmark might eventually reflect what such a system could produce. That could make patents harder to obtain for results that standard AI can readily reproduce. This is a proposed way to adapt the patentability test, not a current, general USPTO standard for assessing inventions.
Disclosure should focus on the invention
The submission also separates disclosure of the invention from disclosure of the AI’s internal process. In its view, applications should meet the same written-description and enablement standards as other inventions: a skilled person should be able to understand and perform the claimed invention. The fact that an AI generated the result, or that a learning system’s internal weights changed during training, would not by itself determine whether the invention is adequately disclosed.
At the same time, the author proposed that applicants disclose when AI contributed to conception. That information could help assess inventorship and provide a clearer picture of how AI is being used in different technical fields. For industrial organizations, consistent records of AI involvement may therefore be useful even as legal requirements and policy approaches continue to develop.
Practical implications for industrial R&D
AI-generated inventions sit at the intersection of engineering practice, patent rights and innovation policy. Companies adopting AI for design, process development or technical problem-solving should distinguish routine analytical assistance from a system’s potential contribution to conception, while ensuring contracts address rights to resulting work. They should also assess patent applications based on the invention’s disclosure and merits, rather than assuming that AI involvement settles eligibility or inventorship.
The broader policy debate remains unresolved. Its practical message for automation leaders is more immediate: as AI becomes embedded in R&D, organizations need clear processes for tracking contributions, managing ownership and evaluating patentability. Those practices can help teams make informed decisions about protecting technical advances while the legal framework continues to adapt.