Charlie Angus’s Substack piece We Demand To Know Our History resurrects the spirit of 1812’s Isaac Brock not as a historical footnote but as a tactical framework for resisting modern digital imperialism, arguing that Canada’s sovereignty in the AI era hinges on reclaiming technological self-determination from foreign-controlled platforms—a thesis gaining urgency as 2026’s AI security architectures reveal systemic dependencies on U.S.-aligned cloud and chip supply chains.
The Brock Analogy: Asymmetric Resistance in the Age of Agentic AI
Angus draws a parallel between Brock’s apply of terrain, Indigenous alliances, and psychological warfare against a superior invading force and today’s need for Canada to leverage its niche strengths—like its AI safety research cluster in Montreal and its critical mineral reserves—to counterbalance U.S. And Chinese dominance in foundational AI models. This isn’t nostalgia; it’s a call for strategic patience, mirroring how elite hackers in the AI era delay exploitation until systemic weaknesses align, as analyzed in recent CMIST research. The analogy holds because both scenarios involve a weaker actor denying an adversary the luxury of predictable outcomes—Brock by making invasion costly, Angus by advocating for Canada to become a “spoiler” in the global AI stack through controlled exports of AI training data governed by Indigenous data sovereignty laws.
Under the Hood: Why Canada’s AI Sovereignty Gap Is Structural
The information gap Angus identifies isn’t just about awareness—it’s about architecture. Canada currently hosts less than 0.5% of global AI training compute, per Stanford’s 2026 AI Index, and relies on AWS and Azure for 89% of its public-sector AI workloads. This creates a latent vulnerability: if U.S. Export controls tighten—as they did in 2025 with the CHIPS Act amendments—Canadian AI startups could lose access to frontier models like GPT-5 or Gemini Ultra overnight. Worse, most Canadian AI models are fine-tuned on U.S.-centric datasets, embedding cultural biases that undermine local policy relevance. For example, a 2026 audit by the Vector Institute found that 74% of Canadian healthcare LLMs failed to recognize French-Canadian medical terminology due to training data sourced primarily from U.S. EHR systems.
Ecosystem Bridging: The Open-Source Counterweight
Here’s where Angus’s call for historical awareness meets actionable resistance: Canada’s underfunded but growing open-source AI ecosystem could become its asymmetric advantage. Projects like FrancoBERT, a French-language LLM trained on Quebecois legal and medical texts, demonstrate how localized models can bypass platform lock-in. Unlike proprietary APIs that charge per token and retain usage data, FrancoBERT runs on modest hardware—achieving 82% accuracy on Canadian French NLP benchmarks using a 7B-parameter model fine-tuned on 12GB of curated text, per its GitHub README. This mirrors the logic of distributed resistance: compact, interoperable units that collectively deny an adversary control over the battlefield.
“True technological sovereignty isn’t about building the biggest model—it’s about controlling the data pipeline and the inference environment. Canada’s strength lies in its ability to govern data at the source, not in chasing parameter counts.”
— Dr. Sasha Luccioni, Hugging Face Climate Lead and former Mila researcher, interviewed April 2026
What So for Enterprise IT: The Mitigation Stack
For Canadian CIOs, Angus’s thesis translates into a three-layer mitigation strategy: First, mandate data residency for AI training using tools like OpenShift CSI drivers to maintain sensitive datasets within Canadian jurisdictional boundaries. Second, adopt hybrid inference architectures—running smaller, fine-tuned models locally via Ollama for edge cases while reserving frontier models for non-sensitive tasks. Third, participate in federated learning initiatives like the Pan-Canadian AI Strategy’s health data consortium, which allows model improvement without raw data leaving provincial servers. This approach mirrors how Brock used militia and Indigenous fighters not to win pitched battles but to disrupt supply lines and erode enemy morale.
The 30-Second Verdict
Charlie Angus isn’t asking Canada to win the AI race—he’s asking it to refuse to run on someone else’s track. In an era where AI dominance is measured in exaflops and token costs, the most radical act may be insisting that your data, your language, and your laws shape the models that govern you. As agentic AI systems begin making autonomous decisions in finance, defense, and healthcare, the lesson of 1812 isn’t about winning battles—it’s about making occupation too expensive to sustain. And in the AI era, that means controlling the compute, the data, and the right to say no.
