Can AI-powered drug discovery earn patent protection in Canada?

AI transforms drug discovery and redefines what innovation means 

Sometimes it really is the journey, not the destination, that counts. This idea captures the current debate over the patentability of AI-enabled drug discovery, where the method itself, rather than the final molecule, may hold the real value. A good example is Alphabet’s Isomorphic Labs. The company has been transforming how drugs are discovered. It has built an integrated AI-driven discovery process that pulls in biology and chemistry data, trains models to identify promising molecules, and lets the AI propose what to make or test next. Each round of experiments adds new data back into the system, refining the model and improving preclinical readiness. 

The Isomorphic Labs process has drawn intense industry attention and serious capital. In January 2024, Isomorphic Labs signed two discovery collaborations with Eli Lilly and Company and Novartis that are potentially worth nearly US$3 billion if milestones are met. In March 2025, the company raised US$600 million to push its AI-designed programs toward first-in-human studies. By July 2025, Fortune and Fast Company were already reporting that Isomorphic Labs was preparing to begin human trials before year-end. 

The drug candidates are valuable, but the way Isomorphic Labs finds them may offer the greater riches. A rival copying the methodology could spin up the same discovery engine and skip years of trial-and-error. That possibility raises the question: can patent law protect the AI-driven drug discovery process itself? 

Patentability depends on the invention, not on who invented it 

 AI-powered drug discovery is, first and foremost, a question of invention, not inventorship. The patentability inquiry asks whether the claimed AI-driven process fits within section 2 of the Patent Act: as a “new and useful art, process, machine, manufacture or composition of matter.” It also asks whether the claims disclose a concrete, reproducible technical process rather than a “mere scientific principle or abstract theorem” under section 27(8). Using purposive construction, the key is to identify what the claims actually stake out. In the context of AI-enabled research, the “scientific principle or abstract theorem” often takes the form of an algorithm or model. The test, then, is whether the process applies that algorithm in a concrete way to produce a specific technical result, or merely describes an abstract computational idea. 

By contrast, whether a patent application can name an AI as an inventor is a separate and narrower issue. Current Canadian law, including the Thaler, Stephen L. (Re), 2025 CACP 8 ruling before the Patent Appeal Board, recognizes only natural persons as inventors, so an AI system itself is not eligible. That does not bar AI-assisted inventions. If a human makes a material contribution to the inventive concept, such as designing the AI workflow, defining the problem, or interpreting the model’s outputs, the application may proceed with that human as inventor.  This aligns with Canadian inventorship doctrine and Canadian Intellectual Property Office (CIPO) practice, and is consistent with the guidance of the United States Patent and Trademark Office, which states that patents may be granted for AI-assisted inventions with a significant human contribution. 

Patentability turns on concrete, reproducible workflows 

AI-powered drug discovery can be patentable when the legal requirements are met. To satisfy utility and enablement under section 27(3) of the Patent Act, the application must describe the process in sufficient detail for a person skilled in the art to reproduce it without undue experimentation. The claims should anchor the method in measurable features, such as how model outputs integrate laboratory automation, what thresholds or triggers advance each iteration, how feedback cycles retrain or constrain the search, and how the system delivers a concrete technical improvement in drug discovery. A bare statement that the AI learns and improves, without those specifics, will read as vague and not reproducible. 

When the legal requirements are not met, an AI-powered drug discovery process is not a patentable invention. Claims usually fail when they are too abstract or vague, and they may succeed only when they describe specific technical operations that are tied to real laboratory or chemical processes. The case law shows how this distinction works. In Amazon.com, Inc. v. Canada (Attorney General), the Federal Court of Appeal held that examiners must first purposively construe the claims and then decide whether, viewed as a whole, the claimed subject matter is merely an abstract idea or mathematical scheme carried out by a computer; if so, it falls outside patentable subject matter. Twelve years later, in Canada (Attorney General) v. Benjamin Moore & Co., the Federal Court of Appeal confirmed that there is no special or rigid test for computer-implemented inventions and instructed CIPO to apply the existing jurisprudence flexibly and focus on whether the claim shows a genuine practical application. The Patent Appeal Board decision in Benjamin Moore & Co. (Re) also helps illustrate what this means in practice: claims that simply describe results-oriented algorithms running on generic computers are not eligible for patent protection, while those that detail the cooperation of technical elements, such as specific instruments, reagents, control parameters, data collection steps, and feedback loops integrated into laboratory automation or chemical synthesis, have a clearer path to eligibility. This is because they move the claim, after purposive construction, from an abstract idea to a concrete, reproducible technical application, as required by the Patent Act section 27(8) and 27(3). Taken together, these authorities show that successful AI discovery patents must connect the algorithmic model to concrete technical operations that produce a measurable outcome, rather than leaving the claim at the level of abstract design or data analysis. 

AI-assisted discoveries face the same core tests as any other patent 

It is easy to assume that an invention loses its originality when artificial intelligence takes part in the inventive process. Yet as AI becomes an inseparable part of modern research, the boundary between human creativity and machine assistance is no longer clear-cut. Canadian patent law does not reject AI involvement outright; instead, it asks whether the invention meets the same foundational requirements as it requires of any other patent application. The question is not whether the inventor used AI, but whether the claimed process represents a concrete, reproducible, and useful application of technology. If those requirements are met, even an AI-assisted discovery can stand as a genuine human invention worthy of patent protection. 

The opinion is the author’s, and does not necessarily reflect CIPPIC’s policy position.