The United States government recently ordered AI company Anthropic to suspend foreign nationals’ access to Fable 5 and Mythos 5, two of its most advanced AI models, citing national security concerns. Anthropic responded by disabling the models for all customers. Organizations in Canada, Europe and around the world that had embedded those tools in their workflows found them simply gone. No appeal process. No migration window. No warning. No jurisdiction over this decision.As the G7 summit wraps up in Evian, France, the Anthropic shutdown has put AI sovereignty and concerns about U.S. dominance high on the agenda. For Canada, the Anthropic shutdown is not just a technology story. It is the bill arriving for past choices — years spent celebrating AI research leadership while under-investing in the commercial, capital and governance conditions to turn leadership into lasting sovereignty. Canada’s new AI for All strategy recognizes that dependence on foreign technology is a strategic vulnerability. It commits billions to compute infrastructure, skills development and domestic AI capacity. These investments are necessary and welcome. But you can build your own data centres and still depend on someone else’s models. Canada needs to build the commercial conditions, long-term investment, procurement pathways and scale-up support that would allow AI companies to grow here rather than leave.Canada’s AI research leadershipCanada’s contribution to modern AI is a matter of historical record.Yoshua Bengio, a computer science professor at the Université de Montréal, helped make Montréal one of the world’s leading AI research hubs. Computer science professor Geoffrey Hinton’s foundational deep learning work at the University of Toronto, which led to a Nobel Prize, was supported by public funding and helped lay the ground for today’s AI systems. Geoffrey Hinton deliveres his Nobel Prize banquet speech in December 2024 at Stockholm City Hall. In 2013, Google acquired Hinton’s startup DNNresearch, which he had incorporated with two graduate students, Alex Krizhevsky and Ilya Sutskever. Both moved to Google. Later, Sutskever co-founded OpenAI, a leading player in the current AI boom. Both Krizhevsky and Sutskever were trained in a Canadian research ecosystem supported by public investment, and both went on to do their most consequential commercial work elsewhere.Canadian researchers remain deeply woven into the systems shaping the global economy. But there is a real gap between scientific leadership and commercial control, and Canada has not closed it. Read more: Canada’s ‘AI for All’ strategy has ambitious growth targets, but it falls short on workers and the environment Sovereignty means accessThe Pan-Canadian AI Strategy, launched in 2017 with $125 million, was celebrated internationally as visionary. But Canada has struggled to turn that research strength into domestic commercial control. Only seven per cent of the IP rights generated through the strategy are owned by Canadian private sector firms. In 2024, two-thirds of high-potential Canadian-led startups that raised more than $1 million were headquartered outside Canada. And Canada deployed less than two per cent of global AI venture capital despite hosting roughly 10 per cent of the world’s top-tier AI researchers. Research leadership alone does not create sovereignty. Sovereignty depends on who owns the companies, controls the models and sets the terms of access. And Canada is still playing catch-up. Strategic dependencyMost Canadian organizations using AI license access from a small number of American companies operating under U.S. law and subject to U.S. government decisions. That relationship looks commercial until a national security order turns it into a strategic dependency overnight.Think of the Hydro-Québec scenario in reverse. Imagine Ottawa ordering Hydro-Québec to stop exporting electricity to the northeastern United States overnight on national security grounds. American businesses and institutions would quickly discover that a routine commercial relationship was also a critical dependency. Canada is now on the receiving end of that lesson with AI.Who decides access?The most advanced AI systems are not nuclear weapons. But they are becoming consequential enough to raise a similar governance question: when a technology can affect economic competitiveness, research capacity, public services and national security across borders, should access be decided by one government and a handful of companies?For now, that is largely how the system works. The U.S. controls an estimated 74 per cent of global AI supercomputer capacity (China controls 14 per cent), while leading American firms build and operate most of the frontier models others rely on.Questions of access, ethics and safety have global consequences, yet they are currently addressed largely through national laws and corporate decisions, and with no genuine consultation or public participation.The Anthropic case makes these questions harder to ignore. Read more: Artificial intelligence raises profound moral questions — for all of humanity to answer A board-level riskNational strategies take years. The Anthropic suspension took hours. That is the kind of disruption organizations should plan for. Guidance from the U.S. Department of Commerce already recommends planning for deactivation or disengagement, including workarounds, alternative processes and clear communication when access changes.Canadian organizations need to treat AI dependency as a significant risk, not an IT purchasing question. That means knowing where AI is embedded in critical workflows, whether those workflows depend on a single provider, what happens if access changes and whether the organization retains enough human expertise to operate without the tool. It means demanding vendor contracts that specify what happens when access is suspended, what data is retained and what notice obligations exist.Canada’s future voiceThe lesson is not to stop using AI. Adoption will continue and the productivity gains are real. The lesson is to stop assuming access will always be there.Canada is now investing billions to strengthen domestic AI capacity. That matters. But sovereignty is not simply about where the servers sit. It is about who controls the capabilities running on them and what happens when that control is exercised by someone else.Canada helped build the present and the future of AI. The question now is whether it is willing to build enough commercial and governance capacity to have a meaningful voice in who controls it. Because whether we have a say or not, we will feel the consequences.Simon Blanchette does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.