Canada hardens physicality test for AI and software patent claims
Canada’s patentability analysis for AI and computer-implemented inventions is moving into a markedly less forgiving phase. CIPO’s March 2026 Practice Notice put the spotlight squarely on the “physicality” requirement implicit in subsection 27(8) and section 2 of the Patent Act: if a claim merely has a computer process an abstract algorithm in a well-known way, without a discernible physical effect, physical change, or an improvement in the functioning of the computer itself, eligibility risk rises quickly.
By late June, practitioner feedback suggests that this is no longer just a policy paper people cite in seminars. Examiners appear to be applying the framework more consistently in day-to-day prosecution, especially for AI models, software rulesets and data-driven diagnostic methods. Saying that a model predicts better, ranks better or diagnoses better is no longer enough on its own. Applicants are increasingly being pushed to show where the physicality lies and why the claimed invention is more than a bare practical application of an abstract idea.
The framework is now more direct, not just stricter
The most important change is not simply that CIPO sounds tougher. The path of analysis has become cleaner and easier for examiners to operationalize. The March 2026 Practice Notice says that earlier MOPOP references to “contribution,” a “technological solution to a technological problem,” and problem-solution reasoning for essentiality no longer apply. Examination starts with purposive construction and then returns to subject-matter under section 2 and subsection 27(8). The real screening work is done by a practical question: as claimed, does the invention have physical existence or manifest a discernible physical effect or change?
That matters because a great deal of AI drafting still relies on functional language that sounds technical but remains abstract at the point where subject-matter is tested. References to servers, processors, storage, inference engines and deployment environments do not do the job by themselves. CIPO’s own text is explicit: the presence of a computer does not necessarily satisfy physicality. If the computer is only carrying out an abstract algorithm in a conventional manner, and there is no improvement in computer functioning or no genuine cooperation with physical inputs or outputs, the claim remains exposed.
Where many AI applications will now run into trouble
The immediate risk point is easy to state and surprisingly common in practice: an application describes a strong commercial or analytical result, but the result is still abstract in the patentability sense. Better recommendations, more accurate classification, sharper fraud detection, stronger medical correlation, improved forecasting — all of these may matter commercially, yet they do not automatically show physicality. Examiners can still ask a blunt question: what changed in the physical world, or what changed in the functioning of the computer as a computer?
For black-box model claims, that question can be fatal unless the specification has already done the hard work. Applicants now need to build one of several clearer pathways into the file. One route is to show an improvement in computer operation itself, such as lower memory use, fewer processing cycles, better scheduling, improved transmission efficiency or another technically grounded performance effect. Another is to anchor the claim to cooperating physical steps, sensors, measurement inputs, control outputs or non-generic hardware interactions. A third route, narrower but still important, is to frame certain diagnostic inventions around concrete testing or measuring steps rather than pure inference alone.
Why Canada currently feels tougher than the United States
From an applicant’s perspective, Canada now often feels less tolerant than the United States for software-heavy claims. Part of that is legal structure, but part of it is administrative clarity. The U.S. debate still turns on overlapping vocabularies — abstract ideas, significantly more, technological improvement, inventive concept-type arguments in practice. CIPO’s latest notice is more concentrated. It places physicality at the centre and uses that concept to sort out what is merely an abstract rule running on a generic computer from what is properly claimed as an invention.
That does not mean Canada has closed the door on digital innovation. The door is still open where the file actually supports the right story. CIPO expressly leaves room where an algorithm improves the functioning of the computer, or where the claimed invention includes meaningful physical cooperation through measurement steps, sensors, robotic outputs or other non-generic elements. The practical difficulty is that this room must be designed into the application from the start. Applicants who rely on a single global English draft and assume Canada can be handled later are likely to find that the support they need is thinner than expected once a subject-matter objection lands.
Diagnostic methods still have room, but only if the physical steps stay visible
One point that deserves closer attention is CIPO’s treatment of diagnostic methods. The 2026 notice does not collapse diagnostics entirely into the same bucket as pure software logic. It preserves a more workable route where the construed claim includes physical steps for measuring, identifying, detecting or assaying the presence or quantity of an analyte in a sample. In that situation, the abstract correlation is not automatically disqualifying, because the claim can still define a practical application that satisfies physicality.
That distinction matters for digital health, AI-assisted diagnosis and biomarker-based platforms. A claim drafted mainly around a score, a probability or a model output may drift toward abstraction very quickly. A claim that keeps the sample-handling, detection, measurement and data-acquisition structure in view stands on firmer ground. This is not a safe harbour, and applicants should not overread it. But Canada is still signalling that a diagnostic invention with real physical testing steps is not in the same position as a software claim that reduces everything to information processing and a result on a screen.
How applicants should adjust now
The first adjustment is drafting order. Instead of starting with business logic, model advantages and use cases, then trying to add hardware language near the end, applicants should begin by asking three practical questions. Where is the discernible physical effect? Does the invention improve the computer or system itself? If not, what concrete physical input, measurement, control or output is genuinely cooperating with the algorithm? If those questions cannot be answered early, the Canadian filing strategy is already unstable.
The second adjustment is to stop treating Canada as a simple downstream translation of a U.S. or EP strategy. The current Canadian issue is not a lack of technical vocabulary. It is the lack of a claim story that makes physicality legible to the examiner. Third, applicants with pending files should review their specifications sooner rather than later. If the support for computer-function improvement, physical measurements, sensor interaction or control outputs is weak, it is better to know that before prosecution hardens. In Canada’s current framework, AI and software patents are still available. What is disappearing is the assumption that a well-packaged abstraction will be enough.
