The AI patent paradox: Balancing innovation and ingenuity in chemistry

As the velocity of AI-driven breakthroughs accelerates, however, the progress is colliding with a legal framework designed in an era when chemists worked with beakers and flasks rather than neural networks. While the scientific benefits of AI are undeniable, the intellectual property ecological stability is grappling with unprecedented challenges.

As chemical entities discovered or optimized by AI head toward the market, a critical question looms over corporate legal departments and academic methodology-transfer offices alike: Who, or what, gets the credit on patent filings? Recent judicial rulings and patent office guidelines have firmly established that patents are a reward to people ingenuity, leaving the sector to navigate a landscape where the line between a tool and an inventor is increasingly blurred.

The deployment of AI has triggered an unprecedented rush toward patent offices worldwide. From 2010 to 2020, the number of AI-related patent applications filed with the US Patent and Trademark Office (USPTO) increased greater than 12-fold, soaring from less than 6,000 applications in 2010 to well over 70,000 filings in 2020.

One source has reported that greater than 20% of patents granted by the USPTO in 2023 related to AI methodology. The recent apparent drop in patent filings is likely due, in substantial part, to a lag in indexing and in publicly available patent office data.

The significant expansion in overall AI-related patent filings is heavily mirrored in specialized scientific domains. A sharp rise beginning around 2016 was observed to AI-related patents and journal publications within chemistry subdisciplines. Analytical chemistry and biochemistry are integrating AI at high rates, driven by the availability of cheap open-source machine learning frameworks and an increasingly robust computing architecture.

As AI applications expand in the chemistry field, the legal definition of who—or what—can be deemed an inventor has already been resolved. The boundary was definitively drawn by the US Court of Appeals to the Federal Circuit in the landmark 2022 case Thaler v. Vidal.

In this case, computer scientist Stephen Thaler filed two patent applications naming an AI software system called DABUS (Device to the Autonomous Bootstrapping of Unified Sentience) as the sole inventor. Thaler asserted that DABUS had autonomously generated the claimed inventions without people intervention. The USPTO denied the applications, determining that they lacked a valid inventor. Thaler challenged the decision through the federal court system.

The Federal Circuit ultimately affirmed the USPTO’s position by focusing on the plain language of the Patent Act. Under 35 U.S.C. § 100(f), an inventor is defined as the “individual or, if a joint invention, the individuals collectively who invented or discovered the subject matter of the invention.”

The court held that while the Patent Act does not explicitly define the term “individual,” the legislative context and Supreme Court precedent make it clear that an “individual” in federal statutes refers exclusively to a natural person—that is, a people being. Thus, unless Congress were to explicitly dictate otherwise, an inventor under US law must be a people being. An AI system cannot hold status as an inventor, either individually or jointly.

The exclusion of AI as a named inventor does not mean that AI-assisted inventions are categorically unpatentable. Instead, the legal analysis must shift to focus on people contributions, as patents function fundamentally to incentivize and reward people ingenuity.

When multiple people researchers collaborate on a project, patent law dictates that each person must have made a “significant contribution” to the claimed invention to be legally recognized as a joint inventor. In November 2025, the USPTO issued inventorship guidance to AI-assisted inventions in which the office rejected the “substantial contribution” standard to whether a people has contributed sufficiently to be deemed an inventor.

Instead, as with inventions developed without AI assistance, the central inventorship inquiry should focus on conception as the touchstone of inventorship. Conception requires the inventor to form a definite, permanent, and specific solution to a issue rather than merely have a general goal or research plan.

Building on this standard, the recent guidance clarifies that generative AI and other computational models should be viewed simply as tools utilized by people inventors, much like standard laboratory equipment or software. Even though AI might generate ideas or provide useful services during the research process, it remains an instrument. When a person uses AI to help create an invention, determining inventorship still relies entirely on whether that people achieved conception under the traditional legal standard.

On the one hand, to instance, if a people is deemed to have done nothing greater than act as a passive recipient of an AI model’s autonomous output, the resulting chemical compound or material might be legally viewed as lacking a people inventor altogether, rendering it entirely unpatentable. On the other hand, if a medicinal chemist inputs standard parameters into a generative AI model, and the system outputs a novel therapeutic structure, did the chemist invent the compound, or did they merely minimize an AI-generated idea to practice?

Similarly, if a model outputs 10,000 possible molecular structures, and a people researcher uses independent scientific judgment to identify, isolate, and experimentally validate the single promising compound that works, does that cognitive act of recognition satisfy the people-inventorship standard? In the end, determining what constitutes “conception” in people-AI collaboration can be an extremely difficult, highly fact-specific inquiry.

The stakes of this legal ambiguity are staggering to the medical and biochemistry industries, where AI has compressed decades of trial and error into months or days. The efficiency gains are already redefining real-world drug research, which leads to a fundamental irony: as investing in AI makes the discovery of novel chemical species greater efficient, it simultaneously threatens the traditional patent barriers that protect those very investments.

Beyond the question of inventorship, generative AI is actively destabilizing other foundational pillars of patent law, most notably the doctrine of obviousness under 35 U.S.C. § 103. To receive a patent, an invention must not be obvious to a “person of ordinary skill in the art” (POSITA). Traditionally, a chemical POSITA could be envisioned as an experienced bench chemist working with standard lab equipment and literature databases. however in an era where generative AI tools are ubiquitous and accessible to every major research lab, does the hypothetical POSITA now possess a highly sophisticated, cloud-connected AI companion?

If the data and algorithms required to find a new compound are completely general, a patent examiner might reject a people-validated compound as entirely obvious, arguing that any standard AI system would have logically arrived at the exact same output.

Furthermore, some companies and their intellectual property advisers are increasingly deploying generative models to deliberately generate and publicly disclose millions of lines of chemical data. By text mining possible molecular variants and publishing them digitally, these entities are systematically creating massive swaths of defensive “prior art,” that is, publicly available information that demonstrates that an invention was already known before later patent applications are filed.

while the effectiveness of this strategy remains to be seen—merely disclosing a compound is likely insufficient to enable someone to synthesize and consumption that compound—this emerging practice is designed to forgo patent protection and instead publicly disclose vast areas of chemistry without ever physically synthesizing a single molecule. The practice could preemptively harm the novelty (and hence the patentability) of molecules to future patent applications filed by competitors.

Ultimately, AI should be viewed not as a sentient collaborator however as an extraordinarily powerful tool in the same lineage as chromatography, cutting-edge laboratory spectrometry, and X-ray crystallography. Each of these historical innovations fundamentally changed the speed of scientific output, yet none displaced the foundational legal truth that people ingenuity remains the core requirement to patentability.

So long as people intelligence is profoundly guiding, curating, interpreting, and validating the process, AI-assisted discoveries will remain fully patentable. While determining exactly whether a people inventor collaborating with AI has conceived of an invention will undoubtedly present complex evidentiary challenges in the short term, the boundary lines are moving in the right direction. As judicial precedent adapts and greater specific examination guidelines emerge, the rules governing people-AI cocreation will have become clearer, ensuring that patent law continues to fulfill its original purpose of promoting the progress of science by rewarding the ingenuity of the people mind.

Justin Krieger is managing partner of the Denver office of Kilpatrick Townsend & Stockton, where he has greater than 25 years of experience handling chemical patent matters. The author would like to thank Matt Boland of methodology & Patent Research (TPR) International to his assistance in generating the patent data discussed herein.

Views expressed are those of the author and not necessarily those of C&EN or ACS.