The Honourable Marjorie Michel, P.C., M.P., Minister of Health,
RE: security concern regarding my personal medical information
Dear Minister.
There is a habit in Canadian health policy of treating data as a resource to be centralized — pooled, deduplicated, and fed upward to Ottawa or a contracted American cloud provider. The instinct is understandable. Population-level insight requires population-level data. The problem is that this logic collides, repeatedly and expensively, with two immovable objects: provincial jurisdiction over healthcare delivery, and the informed privacy expectations of Canadians who increasingly understand what it means to hand over a medical record.
Canada has spent decades trying to build a pan-Canadian health data commons. The result has been a graveyard of interoperability initiatives, stalled by the same jurisdictional fault lines that define Canadian federalism everywhere else. Alberta won't share with Ontario. BC's EHR system doesn't talk to Saskatchewan's. Health Canada has authority over drug approvals and epidemic response but cannot compel provinces to open their clinical databases — nor should it.
Federated learning, the technical core of the Sovereign AI architecture, dissolves this impasse. Under a federated model, provincial hospital networks, GP clinics, and regional labs each train a local inference layer on their own data.
AI models process information directly on edge servers rather than on centralized cloud servers.
What travels to a Health Canada aggregation node is not patient records — it is gradient updates, the mathematical residue of what the model learned. Raw data never moves. Jurisdiction is never violated. And yet a national model accumulates the diagnostic intelligence of forty million patients' worth of clinical encounters.
“The architecture is the jurisdictional compromise. No legislation required.” This resolves a structural tension that has blocked pan-Canadian health AI for two decades — not through negotiation or new legislation, but through technical design. The federated architecture embodies the constitutional compromise that political processes have failed to achieve. Any province that has resisted federal health data sharing on jurisdictional grounds has no objection left to raise.
The connectivity requirement is also inverted. Current cloud-dependent health AI needs a reliable connection to a distant server. Edge inference at the tower runs when the internet goes down — because the tower is the infrastructure. The AI is inside it, not dependent on it.
COROLLARY This is also the pandemic preparedness argument. Decentralized health AI that functions when national networks are degraded or disrupted is a critical resilience asset — directly relevant to post-COVID federal infrastructure reviews. The 2020 crisis exposed how quickly centralized digital health systems fail under unexpected load. An edge-distributed architecture doesn't have a single point of failure.
Canada currently runs its post-market drug safety system largely as a downstream receiver of signals generated by the FDA and EMA. When the FDA flags a drug interaction, Health Canada updates its advisories. When the FDA misses something, Canada misses it too.
A sovereign AI pharmacovigilance layer — trained on Canadian dispensing data, adverse event reports, and provincial EHR systems, operating under a federated architecture with Health Canada as the aggregation authority — would give Canada the capacity to generate its own safety signals before they arrive from Washington. The political risk of epistemic dependence became visible during the Trump-era tariff confrontations and the COVID vaccine procurement scramble: Canada discovered that its regulatory sovereignty was as fragile as its pharmaceutical supply chain.
Canada's existing health data governance frameworks — PIPEDA, provincial privacy legislation, the SCED standards regime — were written with data residency in mind. Unlike US or EU counterparts navigating GDPR retrofits or HIPAA conflicts, Canadian health data law is structurally compatible with federated sovereign AI deployment. Canada is not a disadvantaged jurisdiction here. It is, quietly, a favorable one for pioneering this model internationally.
Thank you for your consideration
John Argast aka Icarus Flyby