Mike Kelly and I continued our work on AI risks and governance issues with an essay in Just Security examining the significance of Canada’s new AI strategy. Here is a short abstract of the essay:
Democratic AI governance is conventionally mapped along a single axis of regulatory intensity, with the European Union’s rights-based AI Act at one pole, the United States’ innovation-first approach at the other, and Canada and Japan occupying a middle position. This essay argues that the axis measures the wrong variable. What matters is not how much a state regulates AI but how it exercises power over AI — whether that power is rule-bound, transparent, and reviewable, or discretionary and exercised at executive speed. Remapped on that basis, the United States, which governs through procurement leverage, export controls, and executive directives rather than statute, sits at the troubling end, while Canada keeps company with the rule-bound regimes.
The significance of Canada’s “AI for All” strategy (June 2026) lies not in splitting the difference between regulation and innovation but in grounding governance in rule-of-law principles — published criteria, mandatory notice, reasons, and recourse — though its reliance on voluntary private-sector compliance remains a serious weakness. The essay further contends that all four models neglect the risk posed by artificial general intelligence, and that only credible ex ante binding commitments can constrain actors against their own competitive interests. Middle powers should emulate the Canadian model pending an eventual treaty framework.
