Last updated on July 25th, 2026 at 05:40 pm
The “Whose AI Runs the Government?” Series: Complete Index
By Jen Evans, Principal, Pattern Pulse AI; co-founder, Tech Reset Canada
Canadian AI sovereignty is Canada’s ability to inspect, govern, constrain, replace and continue operating the AI systems on which its institutions depend. It requires control or enforceable leverage across compute, cloud, models, data, infrastructure, procurement, legal jurisdiction and operational decision-making.
The Five Layers of Canadian AI Sovereignty
- Hardware and compute: Who controls the chips, cloud infrastructure, data centres, networks and energy required to operate the system?
- Models and platforms: Who owns the model weights, software, update mechanisms and analytical architecture—and can Canada inspect, modify or replace them?
- Data and jurisdiction: Where does Canadian data travel, who can access it, and which country’s laws govern or compel its disclosure?
- Operational capacity: Does Canada have the talent, tools and institutional knowledge to deploy, evaluate and maintain the system without permanent vendor dependence?
- Governance and procurement: Do contracts, audit rights, disclosure rules, oversight mechanisms and exit provisions preserve meaningful Canadian control?
AI sovereignty exists only when these five layers work together. Canadian hosting alone cannot create sovereignty if the model, data, operational capability or governing authority remains controlled elsewhere.
The Canadian AI Sovereignty series runs across B2BNN, ResearchGate, and Zenodo. This index collects every piece in one place for citation, sharing, and reference.
NEW:
AI Sovereignty Moves at the Speed of Models, Not Canada
This piece argues that Canada is treating AI sovereignty as a slow-moving infrastructure and procurement project while models, access conditions and strategic dependencies change at release-cycle speed. Sovereignty requires institutions capable of continuously reassessing models, preserving portability and responding before foreign-controlled capability changes underneath Canadian policy.
Updated Stack, July 2026
The Numbered Series
Paper 1: Whose AI Runs the Government? A New Framework for Sovereign AI Assessment (March 1, 2026)
Introduces the assessment framework for sovereign AI, including the five-layer sovereignty gap and the four triggers that turn a procurement decision into a sovereignty exposure: opacity, irreversibility, foreign jurisdictional reach, and institutional capture. Full ResearchGate paper.
Paper 2: The Inverted AI Bubble (March 22, 2026)
Lays out the financial instability thesis for AI hyperscalers, where inverted unit economics, self-consumption loops, and depreciation mismatches make the buildout structurally fragile rather than durably profitable.
Paper 3: The Cost of Dependency (April 7, 2026)
Builds the three-cost-layer structure and the attrition trap, using CoreWeave’s debt profile, the Bell and Saskatchewan data centre analysis, and Ontario’s health records initiative as a live case of Layer 5 exposure.
Paper 4: The Palantir Map: What Canada Has Actually Disclosed (April 19, 2026)
Documents what Canada has actually disclosed about its Palantir footprint and demonstrates the attrition trap in operation, with the Supply Arrangement running to 2028 while the institutional defences meant to constrain it expire first.
Paper 5: The Instruments Canada Needs (April 20, 2026)
Sets out what a functioning sovereignty regime actually requires, the specific legal and procurement instruments Canada lacks, and why their absence leaves the exposure unaddressed regardless of political intent.
Paper 6: May 2026 Canadian AI Sovereignty Update (May 1, 2026)
Updates the series against the month’s developments, tracking how the disclosed contracts and policy signals move against the framework established in Papers 1 through 5.
Paper 7: AI and The Grid: Mythos, Power and Canadian Sovereignty (May 5, 2026)
Examines the intersection of grid security and AI deployment, connecting the National Electricity Agenda to the sovereignty question and showing why energy infrastructure becomes part of the AI exposure surface.
Paper 8: The Coordination Architecture (May 8, 2026)
Maps how the separate pieces of Canada’s AI posture do or do not coordinate across departments, identifying where the absence of a coordinating instrument produces gaps no single actor owns.
Paper 9: The Procurement-Strategy-Policy Gap (May 12, 2026)
Isolates the disconnect between what Canada procures, what it states as strategy, and what its policy actually permits, the same gap a minister revealed when answering a procurement question with a sovereignty deflection.
Paper 10: Palantir: Five Jurisdictions and One Vendor over 48 Hours (April 20, 2026)
Tracks a single vendor’s activity across five jurisdictions in a 48-hour window, illustrating the foreign-jurisdictional-reach trigger as it operates in real time rather than in theory.
Paper 11: Capital Follows Capability (May 13, 2026)
Argues that investment flows toward where capability actually sits, and examines what that direction of capital reveals about where sovereign control is being built and where it is being ceded.
Paper 12: A 2024 Bill C-22 Meets the 2026 Mythos Threat (May 20, 2026)
Tests a 2024 legislative instrument against a 2026 threat environment, showing where the bill was designed for a different problem than the one it now walks into, and what has to change before it passes.
Paper 13: “Moving Forward” Is the Decision (May 28, 2026)
This paper argues that Canada’s latest Palantir procurement is not merely a routine software purchase, but a live data-sovereignty decision already being made through federal purchasing channels. It situates the contract inside a broader convergence of U.S. cloud law, defence procurement, dual-use analytics, and public-sector dependency, showing how “moving forward” can become policy before strategy catches up.
Paper 14: The Maturation and the Convergence: A Canadian AI Sovereignty Series Update (May 28, 2026)
This update frames Canada’s AI sovereignty debate as entering a more concrete phase, where frontier capability, vendor economics, regulatory failure, public-sector procurement, and moral governance are now converging into the same operational problem. It argues that AI’s promise depends on human-centered deployment, sovereign exposure disclosure, and enforceable oversight before systems become too embedded to govern.
Paper 15: Canadian AI Sovereignty: The Exposure CUSMA Cannot Reach
This paper argues that Canada’s AI sovereignty exposure cannot be resolved through the CUSMA review because the real constraint comes from the U.S. CLOUD Act, which can reach Canadian data held by U.S.-domiciled providers regardless of where the servers are located. It calls for Canada to suspend the Section 105 CLOUD Act negotiation, set stronger legal conditions for any future agreement, and build Canadian-controlled compute and cloud capacity so sovereignty means operational control, not just data residency.
Paper 16: The Sovereign Model Is the Missing Layer (June 3, 2026, with Laszlo Lakatos-Hayward, Hyxos)
This paper, co-written with Laszlo Lakatos-Hayward of Hyxos, argues that Canada’s sovereignty debate has fixed on the compute layer while leaving the model layer unowned, and that the model is where sovereignty is actually decided. Using Samsung’s Montreal-built Tiny Recursive Model as the worked counterfactual, Canadian talent and public infrastructure producing a breakthrough whose IP left the country, it shows that a capable model is now buildable at the scale of a national research program rather than a frontier lab. It then sets out six actions in instrument form, led by an edge-first procurement default and a federal IP vehicle, and makes the case for a distributed edge serving layer that can reach service in 18 to 24 months on telecom infrastructure Canada already controls.
Paper 17: Two Documents: A Workplan for a Strategic Problem versus a Vision of the Future (June 4, 2026)
This paper reads the Government of Canada’s AI for All strategy against an Anthropic Institute document published the same week, and argues that the strategy names a goal without naming the constraint that would tell Canada which commitments are decisive. It shows how the plan locates sovereignty in compute and hardware, the layer that conferred AI power in the past, while the frontier has moved to the model and the research loop that runs on it, leaving Canada positioned to host foreign cognition on a sovereign box.
Paper 18: Canada is Building Power Over Speech While Surrendering Sovereignty
Canada’s digital-policy agenda is no longer just a privacy, speech, or cybersecurity debate. Across C-22, C-34, C-36, C-8, and S-209, the federal government is building new powers over platforms, data, identity, encrypted systems, and online expression while Canada’s underlying digital infrastructure remains dependent on foreign clouds, foreign platforms, foreign AI models, and foreign legal regimes. This piece argues that the real sovereignty crisis is not simply what Canada can regulate at home, but what it no longer controls abroad. Canada and the United States are moving through the same digital-policy moment by different routes: expanding domestic authority while surrendering structural independence.
Paper 19: Privacy Legislation Is Pointless If It Does Not Handle Convergence
Privacy law still treats data as if it lives in separate compartments: AI conversations, search history, email, advertising profiles, social graphs, location data, and device data. But AI erases those boundaries. When those layers are fused, they produce a behavioural and psychological model of a person that no single dataset contains and no single regulator is designed to see. This piece argues that privacy legislation organized around individual categories of data is already obsolete unless it governs convergence itself: the act of combining, inferring, retaining, and acting on the fused data object. Canada’s proposed digital-safety and privacy bills may expand oversight, but without a mandate over convergence, they leave the most powerful privacy risk untouched.
Paper 20: Triangulated Sovereignty: The Three Tests Canada’s AI Stack Must Pass at Once (June 19, 2026)
Argues that Canadian government AI now has to pass three sovereignty tests at once, governance under Canadian public-sector expectations, EU-style documentation and high-risk obligations, and credibility as a non-US non-China alternative, and shows how procurement and vendor cost structures install the EU standard inside Canadian deployments years ahead of any statute, extending the dependency model with a fifth trigger: regulatory dependency that ships inside a domestically headquartered vendor. Legislate the standard or inherit it.
Paper 20b: Six Ways a Country Loses an AI System: The Operational Tail of Diplomacy, Trade, and Defence (June 2026)
Pulls all six dependency triggers to the front, infrastructure, supply chain, data, talent, regulation, and interoperability, and argues that AI has become the operational tail of statecraft beneath all three of diplomacy, trade, and defence, using the NATO STANAG history and the alliance’s missing cloud standard to add the sixth trigger: interoperability-shaped dependency, the kind that stays invisible until two sovereign stacks try to share a workload and cannot.
Paper 21: The Whole Stack Is the Lever Now
June 29, 2026
Shows how US control over critical minerals, compute, frontier models and export access is converging into a single trade instrument. It identifies critical minerals as Canada’s clearest source of leverage and argues that sovereignty depends on retaining control of the foundational assets supporting the full AI stack.
Paper 22: The Complete Machine: When Cyber Capability Moves Off the Model
July 1, 2026
Uses Qihoo 360’s Tulongfeng cyber system to show how strategic capability is moving into the operational stack where models, proprietary data and specialized tooling are combined. It extends the sovereignty framework beyond model access and argues that Canadian policy must learn to govern the complete capability loop.
Paper 23: The Two-Year Machine
July 2, 2026
Examines the recurring promise that transformative AI remains one or two years away and the financing, safety and governance narratives sustained by that promise. It adds a fiduciary dimension to Canadian exposure as the American state moves toward becoming regulator, shareholder and strategic gatekeeper of the frontier stack Canada procures.
Paper 24: Sovereignty at the Fabrication Layer: What Canada Can Actually Make
July 5, 2026
Maps Canada’s existing semiconductor footholds across packaging, photonics, compound semiconductors and advanced materials. It argues that targeted ownership and investment can create meaningful leverage without a leading-edge national fab, while showing how insufficient capital allows Canadian-developed capability to migrate abroad.
Paper 25: CUSMA, Canada and Frontier AI: A Zombie Treaty and the State of US Negotiations
July 5, 2026
Uses the nineteen-day US restriction on Anthropic’s Mythos and Fable models to demonstrate that CUSMA cannot secure Canadian access to frontier AI. It proposes workload segmentation, Canadian-controlled infrastructure for essential systems and explicitly revocable access arrangements where foreign frontier capability remains necessary.
Paper 26: Data Adjacency: How Canada Is Now Exposed to AI Systems It Never Procured
July 9, 2026
Introduces data adjacency, the exposure created when Canadian data enters a foreign AI system through shared defence, intelligence or operational environments without Canada purchasing the system itself. NORAD and Project Maven reveal a sovereignty gap that conventional procurement rules, impact assessments and domestic disclosure requirements cannot detect.
Paper 27: The Third Position: Collective Procurement and the NATO Maven System
July 9, 2026
Adds collective procurement to the customer and adjacent positions established in the previous paper. NATO’s Maven acquisition shows how Canada can authorize, fund and operate inside a system while lacking meaningful access to its contract, audit rights, data terms or exit mechanisms, leading to the proposed Algorithmic STANAG.
Media Coverage and Commentary
July 13, 2026 | rabble.ca
Experts warn Carney’s AI strategy will lead to job cuts and environmental damage
Jen Evans discusses Canada’s underdeveloped AI regulation, the absence of disclosed risk assessments, and the need for direct oversight of high-risk technology vendors.
July 12, 2026 | Toronto Star
Canada isn’t ready to assess prisoners using AI, critics warn
Evans examines Canada’s limited capacity to evaluate high-stakes AI systems and the safeguards required before their use in correctional decision-making.
June 8, 2026 | Toronto Star
Carney government testing use of AI in prisons to create profile reports of offenders
Evans explains why probabilistic errors are unavoidable, difficult to detect and capable of propagating through records used for consequential correctional decisions.
June 3, 2026 | Toronto Star
Documents reveal that Palantir contract was worth $30M more than government disclosed
Evans comments on the sovereignty, procurement and accountability implications of the previously undisclosed contract value.
May 26, 2026 | Toronto Star
Ottawa’s latest deal with U.S. data giant Palantir raises warnings
Evans discusses the sovereignty and institutional risks created by Canada’s expanding reliance on Palantir within federal defence systems.
Companion Pieces
Sovereign AI Architecture for Canada: A Visual Walkthrough (ResearchGate, March 2026)
A visual walkthrough of the sovereign AI stack for Canada, showing how the model, compute, and application layers fit together and where control passes out of Canadian hands at each one.
The Sovereignty Word Is Doing All the Work (April 21, 2026)
Examines the SCIP announcement and shows how “sovereignty” was carrying the entire claim while the compute and software layers underneath remained foreign-controlled, with CoreWeave as counterparty risk and no Canadian frontier model in place.
The Temperature Change Consideration Becomes Real: Canada’s Sovereign AI Playbook Just Lost Its Workaround (May 1, 2026)
Marks the point at which Canada’s sovereign AI workaround stopped being viable, developing the temperature change concept as the moment a gradual exposure becomes a structural one.
Fine-Tuning Foundation Models Creates a New Enterprise and AI Sovereignty Risk: Safety Drift (May 1, 2026)
Identifies safety drift as a distinct risk that appears when enterprises fine-tune foundation models, showing how the customization meant to add control can quietly introduce a new exposure.
Canada’s Sovereign AI Triumvirate: What Cohere, CoreWeave, and Palantir Actually Are (May 2, 2026)
Stacks the three counterparties side by side, the model layer, the compute layer, and the application layer, and shows that the sovereignty argument gets harder rather than easier once you name what each company actually is.
Canadian Businesses Face an AI Cybersecurity Paradox: Safer Than the World, But Still Exposed (May 5, 2026)
Sets out why Canadian businesses are safer than the global baseline on AI cybersecurity yet remain exposed, and what that gap means for enterprise risk planning.
The Cohere Command A+ Release: Sovereignty, Speed, Infrastructure, and What Canadian Customers Need to Ask (May 20, 2026)
Examines Cohere’s Command A+ release through the sovereignty lens, including the open-source licensing shift, the infrastructure it depends on, and the questions Canadian customers should ask before deploying.
The AI Procurement Map Just Became Trilateral (May 21, 2026)
Documents the assembly of a third pole in the AI procurement landscape and what the shift to a trilateral map means for Canadian enterprise, government, and defence buyers.
The Model That Owes You Nothing
June 28, 2026
Examines two different economic structures forming around AI: an extraction-based frontier market and an open ecosystem built around shared capability and lower-cost access. It supplies broader economic context for sovereignty by asking how model ecosystems distribute value and what obligations they carry toward the people and institutions that help create it.
Digital Legislative Overreach in Canada Has History and Runs Both Ways
This piece extends the sovereignty series into Canadian digital regulation, arguing that the country’s policy debate has been too narrowly organized around state overreach while missing the deeper exposure created by platform power, foreign jurisdiction, and unpriced access to Canadian markets. Through the Online News Act, Meta’s news blockade, Google’s funding agreement, lawful access legislation, the CLOUD Act, streaming regulation, and the Netflix production commitment, it reframes sovereignty as a question of leverage, recourse, infrastructure, and who gets to extract value from Canadian audiences. t operate beyond meaningful domestic control.
Adjacent Context
Project Glasswing and the Extraordinary Power Paradox We’ve Never Faced Before (April 8, 2026)
Examines a power paradox the governance conversation has not faced before, using Project Glasswing to frame how extraordinary capability outruns the instruments meant to hold it.
The Economic Suicide Schedule: Why AI Chips Die Twice (April 7, 2026)
Explains the double depreciation problem in AI hardware, where chips lose value once to obsolescence and again to the economics of the buildout, feeding the inverted-bubble thesis.
There Are No Watchmen: Why the AI Governance Conversation Answers the Wrong Question (April 12, 2026)
Argues that the AI governance conversation answers the wrong question, focusing on oversight that does not exist rather than the structural exposures that already do.
The U.S. Just Released Its National AI Framework. It Has No Idea What It’s Regulating. (March 20, 2026)
Critiques the U.S. national AI framework for misunderstanding what it is actually regulating, and what that misread means for partners operating under its jurisdictional reach.
When Data Centers Enter the Target Set, Cloud Strategy Becomes Geopolitics (March 6, 2026)
Shows the moment cloud strategy crosses into geopolitics, once data centres become part of the target set and infrastructure decisions carry security consequences.
TSX Venture 50 Signals Canada’s Capital Is Flowing to the Material Backbone of AI (February 24, 2026)
Reads the TSX Venture 50 as a signal that Canadian capital is moving toward the material backbone of AI, the hardware and resource layer, and what that direction reveals.
Jennifer Evans is the founder of PatternPulse AI, and co-founder of Tech Reset Canada. The Canadian AI Sovereignty research and LLM research series are published via Zenodo and ResearchGate at PatternPulse.ai.
