A U.S. export-control directive forced Anthropic last month to suspend access to its two most advanced models for customers worldwide. The U.S. government lifted the controls last week.Patrick Sison/The Associated Press
Michael Geist holds the Canada Research Chair in internet and e-commerce law at the University of Ottawa, Faculty of Law.
For months, questions about digital sovereignty have dominated the Canadian digital policy landscape, with many concerned about domestic control over both computing infrastructure and the data that fuels the digital economy. The debate reflected mounting unease over the risks of relying on non-Canadian companies for what have become essential services, and fears that Canadian privacy safeguards could be overridden by foreign courts or governments.
These remain real concerns, but the past few weeks have revealed an overlooked threat that similarly speaks to a loss of control. While Canadians have been worried about others controlling our infrastructure or using our data, we have lost sight of the risks of Canada being locked out of the most capable artificial-intelligence models, with consequences that could leave the country in the second tier of AI.
Last month, a U.S. export-control directive forced Anthropic, the company behind the popular Claude AI service, to suspend access to its two most advanced models for every customer worldwide, Mythos and Fable. After two weeks of negotiations, the U.S. government lifted the controls last week, restoring the flagship Fable model globally, while the most powerful version remains limited to government-approved organizations. Meanwhile, OpenAI’s newest model is being rolled out, with the U.S. government approving access one customer at a time.
Anthropic calls for global slowdown in AI race to weigh societal implications
The events have positioned the U.S. government as the effective regulator and guardian of new frontier AI models, dictating who has access, when and on what terms. The resolution of the Anthropic standoff entrenches that role, since the company agreed to provide designated U.S. agencies with early access to future frontier models before public release. The launch of a frontier model now involves negotiation with the U.S. government rather than a simple commercial release.
In other words, the U.S. can switch AI access on and off at its discretion, and Canada has no assurance or guarantee of access to models critical to addressing cybersecurity risks and maintaining economic competitiveness. This vests enormous power in a single government and turns otherwise commercial decisions into leverage that can be wielded with few limits.
Canada’s AI strategy implicitly acknowledges the risk by supporting national AI champion Cohere Inc., and backing open models that no single government can control. Both steps are grounded in reducing our reliance on the United States. But the harsh reality is that Canada is not going to outspend or outbuild the U.S. AI giants, and relying solely on open models is not an option for many organizations.
Opinion: Canada’s national AI strategy is an incredible step forward
While there is no surefire way to gain preferential access to frontier AI models, incorporating the issue into our trade negotiations offers the cleanest mechanism. Last week, the U.S. confirmed that it would not renew the United States-Mexico-Canada agreement, triggering what could be years of negotiations over the agreement’s future.
Canada has rarely entered these negotiations with digital issues as a top priority. If anything, Ottawa has caved to U.S. pressure on issues such as the digital services tax and mandated payments by internet streaming companies. Yet now is the time for a digital demand of our own: legal assurances that Canadian firms will be judged against the same security requirements as U.S. companies with no exclusion based on nationality, and failing that, a place at the front of the line for frontier AI models that puts us ahead of other allies.
Artificial Intelligence Minister Evan Solomon looks at Prime Minister Mark Carney as he announces a new federal AI agenda, in Toronto, in June.Cole Burston/Reuters
Canada is already treated as a preferred ally on security matters, sharing intelligence through the Five Eyes alliance of anglophone countries and defending the continent through the North American Aerospace Defense Command, or NORAD.
Canada’s cyber defence agency was also granted access to Anthropic’s most advanced model through the company’s trusted-partner program before the shutdown, with AI Minister Evan Solomon touting the access as essential to protecting Canadian institutions. Excluding Canada from frontier AI while relying on it for joint intelligence would undermine the security of both countries.
Minister Solomon confirms Canada now part of Anthropic’s Mythos AI preview program
There are no perfect options available: Relying on China-based AI is a non-starter, aligning with U.S. security and AI regimes comes at a cost to our regulatory independence, and a middle-power approach risks becoming a middling power in AI.
The sovereignty debate has to date been preoccupied with keeping foreign governments and agencies out of Canadian data. However, access to frontier AI models will play a key role in determining whether our companies can compete and our critical systems can be defended. That is worth fighting for, and the trade table is the best place to do it.