Can You Patent Your AI? New Guidance in Canada and the US for Software and AI Inventors

Why It Matters

The divergent standards dictate distinct claim‑drafting strategies, making eligibility the decisive factor for protecting AI innovations and maintaining competitive advantage.

Key Takeaways

• CIPO requires physical effect or computer improvement for AI patent eligibility
• USPTO's 2025 guidance shifts burden to examiner, favoring AI applicants
• Adding tangible hardware to AI claims restores patentability in Canada
• SMED declarations provide factual evidence to overcome US §101 rejections
• US now accepts broader technical improvements, not just computer performance gains

Pulse Analysis

AI‑driven inventions are now a hot‑ticket item in global patent portfolios, yet the legal landscape remains fragmented. In Canada, the latest CIPO Practice Notice tightens the "physicality" gatekeeper, insisting that a claim either produce a discernible physical change or demonstrably enhance computer performance. The notice also discards the outdated "actual invention" step, aligning analysis with purposive claim construction and recent Federal Court rulings. For AI developers, this means embedding sensors, actuators, or other hardware into claims—or explicitly tying algorithmic advances to measurable computer‑level gains—to survive the eligibility hurdle.

Across the border, the USPTO has taken a more permissive stance. The August 2025 memorandum and the *Ex parte Desjardins* decision instruct examiners to apply the Alice/Mayo test only when the claim clearly recites a judicial exception, and to favor applicants when the improvement is technical, even if it lies outside pure computer performance. The introduction of Rule 132 Subject‑Matter Eligibility Declarations (SMEDs) equips applicants with a factual toolkit—benchmark data, performance metrics, and implementation details—to rebut presumptive rejections. This shift reduces uncertainty and accelerates prosecution for AI‑centric inventions, especially those that improve model training, reduce catastrophic forgetting, or optimize parameters.

Practically, innovators must now tailor their patent strategies to each jurisdiction. In Canada, claim drafts should spotlight physical embodiments—robots, IoT devices, or manufacturing equipment—that execute the AI logic, thereby satisfying the physical‑effect test. In the United States, emphasizing concrete technical benefits and preparing robust SMEDs can tip the balance in close calls. As courts and patent offices continue to refine AI policy, staying ahead of these nuanced eligibility criteria will be essential for securing enforceable rights and preserving market leadership.