Canada Isn't Trying to Build GPT-6. It's Trying to Write the Rules for It.

In October 2024 the Nobel Prize in Physics went to two people for work on neural networks. One of them, Geoffrey Hinton, had spent decades in Toronto. Canada could plausibly claim to have helped invent the technology now reordering the global economy, and it had remarkably little to show for it commercially. The talent trained here; the trillion-dollar companies were built somewhere else, mostly to the south.

So Canada is making a different bet. Not that it will build the next frontier model — it won’t — but that it can become the jurisdiction the democratic world trusts to set the norms, standards and guardrails for artificial intelligence. Call it AI governance as foreign policy: an attempt to convert research prestige into the softer, stickier kind of influence Canada once wielded through peacekeeping.

The research it can’t quite cash

The raw material is genuinely world-class. Mila in Montréal, the Vector Institute in Toronto and the Alberta Machine Intelligence Institute form one of the densest clusters of AI research anywhere. Hinton’s Nobel, shared with John Hopfield, sits alongside Yoshua Bengio’s Mila — co-funded by Quebec, Ottawa and industry, and one of the more successful models of public-private research the country has produced.

The problem is what happens next. The Vector Institute’s own estimates suggest a majority of its graduates take jobs at U.S. companies. Canada is, by some measures, the world’s fourth-largest beneficiary of AI investment after the United States, United Kingdom and China, and it has real commercial players — Shopify in AI-powered commerce, Cohere building enterprise language models, Ada in customer-service AI. But the gravitational pull of American capital and salaries is relentless, and the research heritage keeps leaking downstream. A country cannot easily build national economic power on an industry whose best people it trains and then exports.

The pivot to rules

If you cannot win the capacity race, you can try to win the governance one — and here Canada moved early. Its Voluntary Code of Conduct on advanced generative AI, signed by major developers in 2023, gave the country first-mover status on industry self-regulation, for whatever self-regulation is worth. Canadian officials have shown up as conveners at the AI safety summits — Bletchley Park, Seoul, Paris — positioning the country alongside Britain and France as a standards-setter rather than a mere participant.

The human center of this is Bengio, who has become one of the most cited voices in AI safety worldwide, serving on the UN’s AI advisory body and testifying before the European Parliament. When a Montréal professor is helping shape how the EU and the UN think about frontier risk, that is influence of a specific, Canadian kind — quiet, expert, disproportionate to the country’s size. Domestically, the less glamorous work matters too: federal agencies like the Canada Revenue Agency and Statistics Canada are developing AI-deployment standards that other G7 governments are watching, because someone has to figure out how a democratic state actually uses these tools without wrecking public trust.

Whether it counts as power

Convening summits and cited experts are real, but they are not obviously the same as power, and Canada has a habit of confusing the two. The peacekeeping brand, the comparison everyone reaches for, eventually revealed its limits: a reputation can outlast the capacity that earned it, and Canada spent years trading on a peacekeeping identity long after it had stopped doing much peacekeeping. The risk of an AI-governance brand is identical — that it becomes a story Canada tells about itself rather than an instrument that changes outcomes.

There is a more concrete version of the ambition that would be harder to fake. It is the idea of an “AI CIDA” — a governance-exporting institution, echoing the old Canadian International Development Agency, that would help ASEAN states, African governments and EU partners build their own AI regulatory frameworks using Canadian expertise. That would be governance as a genuine export, not just a conference presence. Nothing like it exists yet.

What to watch

Two tests will show whether the bet is real. The first is whether Canada passes actual domestic AI legislation in 2026, after the Artificial Intelligence and Data Act stalled in Parliament — it is difficult to sell the world on guardrails you have not managed to legislate at home. The second is Cohere: whether Canada’s most credible independent model company grows into a commercial champion, or becomes another promising firm that gets absorbed into the American orbit.

Reading list