Who Owns a Machine’s Idea?
Artificial intelligence has completely transformed the process of invention, enabling the generation of new technological solutions, drug compounds, and engineering designs with minimal human direction. Traditional patent law, however, was built on the assumption that invention is a fundamentally human act, requiring a natural person to conceive it. As AI systems operate with greater autonomy, the human-centered framework of patent law no longer applies. Patent law should be reformed to address fully autonomous AI-generated inventions by creating a limited, specialized form of protection that preserves innovation while preventing excess concentration of patent power to maintain fairness.
The structure of American patent law makes clear how central the human inventor has always been to this country. In fact, the opening line of the Intellectual Property Clause confirms America’s intent to “promote the progress of science and useful arts”. This clause has traditionally been understood as an incentive system. By offering inventors a temporary monopoly, the government encourages people to disclose new, beneficial ideas to the public instead of keeping them secret. The Patent Act builds on this by requiring that an invention be patentable subject matter, novel, and described clearly enough for others to understand. Beyond these technical requirements, there is an assumption that an inventor is an “individual,” meaning a real human. Courts have repeatedly reinforced that inventorship depends on mental conception, which is defined as the formation in the mind of the inventor of a definite and permanent idea. That definition works effectively when dealing with human researchers, but becomes much more complex to apply when the “mind” responsible for an idea is AI.
It is important to recognize that not every use of AI creates this problem. In many research settings, AI acts as a powerful tool. A scientist might use artificial intelligence software to analyze patterns in large amounts of data or to simulate chemical reactions. In those situations, the human researcher would still be the one framing the question, interpreting the output, and ultimately proposing an idea. The AI assists, but it does not necessarily replace human mental conception. The human mind remains central to the creative process. It becomes difficult to apply patent law when AI operates with a high level of autonomy and is the entity responsible for the conception of an idea or invention. Systems trained on enormous quantities of data will be able to generate new designs or compounds without specific human direction. The human may train and design the system, but if the system itself identifies the specific invention, then the human cannot be seen as responsible for it. When no real person forms the inventive idea in their mind, the legal requirement of conception becomes nearly impossible to truly satisfy.
Under current law, courts have made it clear that only humans can be inventors. Attempts to categorize an AI system as an inventor have been rejected because artificial intelligence cannot be classified as an individual. As a result, fully autonomous AI-generated inventions are left in a gray area. If a human cannot truthfully say that they have conceived the invention, then technically, no patent should be issued, even if it meets the other requirements of a patent. This raises serious concerns for companies that invest heavily in AI research. Developing advanced AI systems requires enormous amounts of time, research, and money. If the outputs of those systems cannot be legally protected, firms lose the incentive to invest in them. This will especially occur in high-risk industries like pharmaceuticals, where patent protection is essential for recovering the massive costs of research. On the other hand, if companies can simply name a related human supervisor or programmer as the inventor even when that person did not truly conceive the invention, the integrity of patent law is weakened. The requirement of mental conception would be reduced to fiction.
Some argue that the solution is simple and that AI-generated inventions should receive full patent protection just as a human-created one would. At first glance, this seems fair because the inventions themselves can be highly valuable and deserve to be legally protected. However, this approach would ignore the scale at which AI can operate. A human inventor would realistically produce a limited number of patentable ideas over a lifetime. An autonomous AI system, on the other hand, could potentially generate hundreds of thousands of patentable outputs in a relatively short period of time. If each of those outputs receives a full patent, large corporations with access to advanced AI infrastructure could quickly accumulate massive patent portfolios. This concentration of patents would create barriers for smaller companies and startups that cannot afford similar systems. Instead of promoting progress, the patent system would reinforce the dominance of large companies.
For these reasons, the best solution is not to deny protection entirely or to grant full traditional patents, but to create a limited and specialized form of protection specifically for fully autonomous AI-generated inventions. The first step in this reform would be clearly distinguishing between inventions genuinely conceived by humans and those that are generated through AI. If the inventive step comes directly from the autonomous operation of an AI system, then the invention should fall into a new category with different rules. These inventions could receive a reduced duration of time on their patents, which would reduce the long-term impact of large-scale patent accumulation, in turn reducing competition in America. It would give companies the chance to benefit from their investment in developing AI systems. Transparency is the key to this reform being successful. Applicants seeking protection for their AI-generated inventions should be required to disclose that the invention was produced by a mechanical system and explain how the system contributed.
In the end, artificial intelligence challenges one of the most basic assumptions of patent law, which is that the invention conceived must be a product of the human mind. Ignoring this change could leave valuable innovations unprotected, and companies wouldn’t receive fair returns for their investment in AI. At the same time, granting full traditional patent protection could lead to extreme concentration of intellectual property in the hands of a few powerful companies. A limited and specialized system for fully autonomous AI-generated inventions offers a balanced solution. It respects the original purpose of patent law, which is to promote progress, while adapting to a modern technological reality that the initial drafters of the law could not have imagined would occur.
Bibliography
Artificial Intelligence and Patent Law | congress.gov | Library of Congress. Accessed March 1, 2026. https://www.congress.gov/crs-product/LSB11251.
Media, Menu SLS |News &, and Monica Schreiber. “Patent Disclosures in the Age of Artificial Intelligence.” Stanford Law School. Accessed March 1, 2026. https://law.stanford.edu/press/patent-disclosures-in-the-age-of-artificial-intelligence/.
“Is the Law Playing Catch-up with Ai?” Harvard Law School, January 16, 2025. https://hls.harvard.edu/today/is-the-law-playing-catch-up-with-ai/.
Ahmed Mohamed, Nashwa, Munira Mohamad Salem Rahma, and Mohamed Al Hemairy. “Challenges and Prospects in Granting Patent Rights to Ai-Generated Inventions: A Global Legal Study with Special Reference to the UAE.” Nature News, November 27, 2025. https://www.nature.com/articles/s41599-025-06121-y.
“Patent Basics.” United States Patent and Trademark Office, January 22, 2025. https://www.uspto.gov/patents/basics.
Artificial Intelligence can’t patent inventions: So what? Accessed March 1, 2026. https://www.fbm.com/publications/artificial-intelligence-cant-patent-inventions-so-what/.
“Ai Is Flooding the Zone with Patents. How Can They Be More Reliable?” Duke University School of Law, May 30, 2025. https://law.duke.edu/news/ai-flooding-zone-patents-how-can-they-be-more-reliable.