The Real Danger of AI Developers Becoming Unwanted Co-Inventors

Under current law, only natural persons can be inventors on a patent. But as AI systems become more sophisticated and domain-specific, questions emerge about whether the creators of such AI might contribute to the conception of inventions generated with their tools. Consider this scenario: a scientific researcher uses a highly specialized AI model (designed for, say, molecular drug design) to discover a new pharmaceutical compound. The human runs the model, evaluates outputs, and files a patent application claiming the new compound. Is the AI’s developer – who trained and fine-tuned the model to solve such molecular design problems – a silent joint inventor of that compound? If so, that developer would be a co-owner of the patent by law, even though they never met the researcher. This scenario implicates a looming tension between traditional inventorship doctrine (focused on human inventors) and the evolving reality of AI-assisted innovation. The issue is a cautionary tale: absent clarification, we may soon see AI platform providers unexpectedly claiming inventorship (and ownership) stakes in others’ inventions, solely because of how their AI systems shaped the inventive output.

Patent Inventorship

Under the Patent Act, only a natural human being—can be an inventor. In Thaler v. Vidal, the Federal Circuit confirmed this, holding explicitly that AI systems cannot qualify as inventors. While the court did not rule on inventions created with AI assistance, it emphasized that the Patent Act provides no basis to include non-human inventors. Misidentifying inventors, including listing non-humans, can render a patent invalid or ungrantable.

Joint Inventorship ‘Significant Contribution’ Test

Joint inventors must each significantly contribute to the conception of at least one claimed element of an invention, although they need not contribute equally or work physically together. Under Pannu v. Iolab Corp., co-inventors must: (1) significantly contribute to the conception or reduction to practice; (2) provide contributions meaningful relative to the invention as a whole; and (3) offer more than mere explanations of known concepts. In short, co-inventors must provide non-trivial mental contributions to the invention’s conception.

U.S. Patent and Trademark Office (USPTO) Inventorship Guidance for AI-Assisted Inventions

In February 2024, the USPTO issued formal guidance addressing inventorship in the context of AI-assisted inventions [88 Fed. Reg. 10043 (Feb. 13, 2024)]. This guidance reaffirms that only natural persons can be named as inventors or co-inventors, even if an AI was instrumental in the invention process. At the same time, it clarifies how to apply the “significant contribution” test in scenarios involving AI. The focus must remain on human contributions: if a natural person made a significant contribution to the conception of the invention, that person can be an inventor notwithstanding the use of AI. Conversely, if the inventive contribution came from the AI’s output without significant human creative input, then no patent can be obtained (because there is no qualifying human inventor). The USPTO emphasizes analyzing inventorship on a case-by-case, claim-by-claim basis: every claimed invention must have at least one human inventor who conceived it.

Notably, the 88 Fed. Reg. 10043 (Feb. 13, 2024) provides illustrative principles for AI scenarios. For example, merely using an AI tool or owning/controlling it does not by itself confer inventorship. A person who simply poses a general problem to an AI or recognizes the AI’s output as valuable is not necessarily an inventor if they did not contribute to the conception of that output. On the other hand, a natural person who “develops an essential building block” that leads to the AI-generated invention can be considered to have made a significant contribution to conception. In particular, “the natural person(s) who designs, builds, or trains an AI system in view of a specific problem to elicit a particular solution could be an inventor, where that designing, building, or training of the AI system is a significant contribution to the invention created with the AI system.”  This principle directly implicates developers of specialized AI: if their work in creating the AI was done with a specific inventive outcome in mind (and that outcome is realized in a patentable invention), they may qualify as co-inventors under the traditional test of significant contribution. The guidance underscores, however, that simply maintaining “intellectual domination” over an AI or providing it with resources is insufficient – there must be a concrete contribution to the inventive concept itself. In sum, the rules can be distilled as follows:

• Only humans can be inventors (no AI “inventors”), but…
• Use of AI tools does not disqualify human inventors, so long as the human’s contribution meets the significant contribution threshold.
• Joint inventorship may encompass AI developers if their actions in creating or configuring the AI amount to contributing an inventive idea to the final invention.
• Mere ownership or oversight of AI is not inventorship; a genuine inventive contribution is required from the person claiming to be an inventor.
• These rules, grounded in statute (35 U.S.C. §§?100, 115, 116), case law (Thaler, Pannu, etc.), and USPTO policy, frame the analysis of our core issue.

General AI Tools as Non-Inventive Aids

Under the current legal framework, widely-used AI systems like large language models (e.g., OpenAI’s ChatGPT or Google’s Gemini) are treated as advanced tools or aids in research – akin to software, textbooks, or microscopes – rather than inventive entities in their own right. The human user’s role remains paramount. If a researcher uses a general-purpose AI to help draft a document, compile data, or even suggest ideas, the law views this similar to using any non-inventive aid. The human who recognizes and refines the idea is the one potentially contributing to conception, not the tool. The Federal Circuit explicitly noted that its bar on AI as an inventor does not imply that inventions made with AI assistance are unpatentable. Rather, patentability and inventorship hinge on a human’s inventive input. The USPTO’s guidance confirms that using AI in the inventive process “does not negate the person’s contributions as an inventor” so long as the person indeed made a significant inventive contribution. In practice, this means that a scientist employing a general AI platform is treated no differently than one using any other sophisticated instrument – the inventive spark must come from the human mind. For example, if a chemist asks ChatGPT to suggest new chemical compounds and the chemist, using their expertise, selects one promising suggestion, refines it, and conceives its practical application, the chemist is the inventor. ChatGPT’s developers (OpenAI) would have no claim to inventorship on the resulting patent because the AI was used as a generic tool; OpenAI did not specifically conceive any element of the chemist’s particular invention. Indeed, principle #5 of the USPTO’s AI inventorship guidance makes clear that simply providing or owning an AI system, without more, confers no inventorship rights. In short, general-purpose AI models are presently viewed as non-inventive aids – powerful extensions of human capability, but with no independent claim to inventorship and no automatic inventorship rights accruing to their developers.

The Rise of Purpose-Built, Domain-Specific AI Models

A more complex situation arises with highly specialized AI systems designed ex ante to perform inventive tasks in specific domains (e.g., drug discovery, new material design, engineering solutions). Unlike general AI, these are purpose-built by their developers with a clear inventive goal in mind. We are witnessing an accelerating trend: major AI companies and research institutions are actively building AI models explicitly aimed at solving creative R&D problems and generating patentable solutions. For instance, OpenAI has partnered with pharmaceutical companies to develop custom AI models for drug development, providing “tuned models” and purpose-built solutions for discovering new medicines. Google’s DeepMind unit is pushing the envelope in algorithm and materials discovery – its latest AI agent AlphaEvolve (powered by Google’s Gemini model) autonomously designed novel algorithms that surpassed long-standing human-devised solutions (including a new matrix computation method better than a 50-year-old algorithm). DeepMind has also used AI to discover improved algorithms for real-world problems like data center scheduling and chip design, effectively generating innovations that no human had conceived before. In the life sciences, researchers are deploying large AI models to invent new biological molecules: for example, one of the largest protein-language models ever developed was recently used to create entirely new fluorescent proteins not found in nature. These examples illustrate that AI is no longer confined to rote assistance – in specialized contexts, AI systems are producing creative outputs that qualify as potential inventions. And importantly, those systems are intentionally constructed by their developers to do so.

If the creators of such an AI have effectively conceived part of the inventive idea through their design of the AI, they might arguably be co-inventors of resulting inventions. The 88 Fed. Reg. 10043 (Feb. 13, 2024) directly speaks to this: a person who “designs, builds, or trains” an AI in view of a specific problem to elicit a particular solution could be deemed an inventor if that AI development work was a significant contribution to the eventual invention. This scenario goes beyond using AI as a neutral tool – here, the AI’s developers embed inventive insight into the system. For example, imagine a materials science research group at a tech company like Google or Meta that specifically engineers an AI model to design novel battery materials with higher energy density. They train the model on electrochemical principles and optimization criteria such that the model outputs new candidate compositions. Now, a university scientist, who had no hand in building that AI, uses it and identifies a groundbreaking new battery compound suggested by the model, and files a patent on that compound. The question is whether the AI’s developers have a rightful inventorship claim. They were not in the lab with the scientist, but they did create the “essential building block” – the trained model and its problem-specific heuristic – that yielded the invention. According to the USPTO’s guiding principle #4, this could indeed be a “significant contribution to the conception of the claimed invention,” even though the AI developers were not physically present or directly participating in the experiment. In legal terms, the AI developers provided part of the conception (the idea of a solution approach) by virtue of how they structured the AI.

As of now, the Federal Circuit insists that only natural persons can be inventors. This means an AI itself can never be an inventor, and by extension an AI’s owner or maker is not automatically an inventor either. Joint inventorship requires each co-inventor to personally contribute to the inventive idea. The novel twist with purpose-built AI is determining whether the AI’s makers have personally contributed to the specific invention’s conception, or whether their contribution was merely to create a generally useful tool. The distinction may hinge on how specific and directed the AI’s training was. If the AI was trained generally (e.g., a broad chemistry model that can propose many molecules), its developers are analogous to authors of a reference work – they likely haven’t conceived any particular downstream invention. But if the AI was targeted to solve a particular problem or class of problems (e.g., “design a molecule to inhibit enzyme X” or “find a structural design for component Y”), then the act of building that targeted solution could cross into inventive territory. In joint inventorship doctrine, conception is the “formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention.” If part of that formation occurred in the minds of the AI’s creators (embodied in the AI’s architecture or training weights), patent law might recognize them as co-inventors – provided a human inventor still exists to satisfy the natural person requirement. The 88 Fed. Reg. 10043 (Feb. 13, 2024) implicitly acknowledges this possibility by instructing examiners to consider whether the AI’s designer provided an “essential building block” of the invention. It even suggests that inventorship can be established without a person participating in every step of conception, so long as their contribution (here, the AI’s design) was crucial.

While general AI use poses little inventorship controversy, purpose-built AI systems change the equation. They introduce the possibility that an AI’s developer has contributed an inventive idea at one remove. The current framework would treat that developer as a co-inventor if the contribution is significant to conception. Thus, paradoxically, AI platform providers could start acquiring co-ownership stakes in inventions they did not explicitly set out to make, simply because their AI was designed to invent. This is not a merely theoretical concern – as noted, leading companies (OpenAI, Google, Meta, and others) are heavily investing in domain-specific AI with exactly this inventive capacity in mind. Each new benchmark these systems hit (designing a new drug molecule, discovering a new algorithm, creating a new material) brings us closer to patent applications that will raise these inventorship questions. The trend line is clear: AI is moving up the inventive chain from aid to co-creator. Our laws and policies will need to adjust to this reality to provide clarity and fairness.

Cautionary Outlook

We are quickly approaching a scenario where developers of AI platforms could unexpectedly become co-owners of patents they did not intentionally invent. Under current doctrine, if creating a specialized AI contributes significantly to the conception of a downstream invention, the developers may automatically gain co-inventor status and thus co-ownership. Imagine investing substantial resources into developing a patentable innovation using an AI tool, only to later discover the tool’s provider must be named as a joint inventor, instantly gaining equal patent ownership. Such “surprise” ownership stakes challenge the traditional notion that patent rights derive directly from the inventive acts of identifiable persons. Current inventorship doctrine, created for human collaborators, lacks a clear framework for layered contributions involving autonomous problem-solving tools like AI. This gap leads to uncertainty, hidden legal risks, and potential litigation surrounding high-value, AI-assisted inventions.

At minimum, the USPTO’s recent guidance (while a good start) should evolve into more formal rules after public comment, to give both examiners and innovators clearer criteria. The Thaler case and the USPTO have drawn a bright line that an AI cannot be an inventor, which is sensible and likely to remain the law. But the gray area lies in human inventors who are one step behind the scenes – the AI’s creators. The patent system must provide clarity: are these people inventors, or not, and based on what standard? Without clarity, we risk chilling collaboration and adoption of AI tools (researchers might fear using powerful AI from third parties if it could entangle ownership) or, conversely, unfairly depriving AI developers of credit where it is genuinely due.

Image Source: Deposit Photos
Author: yacobchuk1
Image ID: 162843428