Sovereignty Is Built, Not Banned: Why Bill C-36 Reaches for the Wrong Hand

Sovereignty Is Built, Not Banned: Why Bill C-36 Reaches for the Wrong Hand

By Prof. Barry Appleton · June 16, 2026 · Appleton’s Clause & Effect

On the morning he left for the G7 in France, the President of the United States told the French President to repeal his country’s three percent digital tax or watch one hundred percent tariffs fall on every bottle of French wine and champagne sold into the American market.1 The American market takes close to a fifth of French wine exports, worth more than two billion dollars a year.2 The threat was not improvised. It restored the same 100% rate that the Office of the United States Trade Representative first proposed in 2019, and it followed a presidential directive to the USTR and the Treasury to decide whether to reopen a formal investigation into the French levy.3

France is being singled out because it did not fold. Canada folded. Ottawa walked back its own Digital Services Tax in 2025 after Washington suspended trade talks, and the levy was formally repealed the following year.4 Without Canada, there are 32 countries left with Digital Sales Taxes. Italy is reported to be weighing repeal. The United Kingdom has kept its tax under the terms of its own understanding with Washington, for now.5 The lesson the United States is teaching, in public, on the eve of a summit it was attending as a guest, is simple. A domestic digital measure it dislikes will be answered with tariffs on something unrelated and politically painful. Wine for a French tech tax. Lumber, steel, or dairy for a Canadian one.

The tariff is only one of the two instruments now in play. The other is access itself. On June 12, 2026, the United States Commerce Department ordered Anthropic to cut off its two most capable AI models, Fable 5 and Mythos 5, for every foreign national, inside the country and out, including the company’s own foreign-born staff. Anthropic could not sort its users by passport in real time, so it disabled both models for everyone on the planet in order to comply.6 A tool that hundreds of millions of people had used days earlier was dark by the weekend, on the authority of a government that no foreign user had any standing to question. The pattern now shows in both instruments. Where Canada regulates American platforms, Washington answers with tariffs. Where Canada depends on American infrastructure, Washington holds the switch. One instrument punishes what Canada does. The other decides what Canada can do at all.

Canada noticed at the very top of the food chain. Speaking to reporters in Ireland on June 14, two days before the G7 took up artificial intelligence, Prime Minister Mark Carney drew the lesson in plain terms. The episode, he said, “is something that can happen with overreliance on certain models.” Then came the line that belongs above every desk where Canada’s digital laws are written: “Nobody has done anything wrong in the situation. But we will have done something wrong if we just accept this, don’t take the lesson, don’t build out and diversify.”7 A former central banker was describing model risk, the danger of resting an entire system on a single point of failure. The lesson he named was construction. Build out. Diversify. Not restriction.

“Nobody has done anything wrong in the situation. But we will have done something wrong if we just accept this, don’t take the lesson, don’t build out and diversify.”

Prime Minister Mark Carney, Ireland, June 14, 2026

I have traced the anatomy of that second instrument before in this space: see “The Digital Hinge of Sovereignty,” and “The Cloud Casts a Long Shadow,” .

This is the board on which Parliament has just placed Bill C-36, the Protecting Privacy and Consumer Data Act, tabled on June 15 by the Minister of Artificial Intelligence and Digital Innovation as a cornerstone of the government’s “AI for All” strategy.8 The bill is the most consequential rewrite of Canadian private sector privacy law in a generation. Its goals are sound. Its method, in several places, is the one method the 2026 CUSMA review is most likely to punish.

What the bill sets out to do

Bill C-36 repeals the operative part of Canada’s existing privacy statute and replaces it with a new Act. It recognizes privacy as a fundamental right. It requires meaningful consent and plain language explanations. It treats children’s data as sensitive by default. It curbs surveillance pricing, the practice of using personal data to charge one consumer more than another. It creates a data mobility framework that enables Canadians to move their information between organizations. It builds a new regulator, the Digital Safety and Data Protection Commission of Canada, with the power to levy penalties as high as twenty-five million dollars or five percent of global revenue. 9 It is enforced by the same Commission that will administer the Safe Social Media Act, the online safety bill tabled five days earlier. 10

None of those objectives is wrong. Every one of them responds to a real harm. The problem is not the destination. The problem is that the bill was drafted inside one rulebook and will be judged under two.

The bill was drafted inside one rulebook and will be judged under two. The first is Canadian public law. The second is enforced in Washington, and its penalties do not fall on the law. They fall on Canadian exports.

The second rulebook

The first rulebook is Canadian public law: the Charter, administrative law, and the division of powers. Officials in Ottawa know it well, and the bill reflects that fluency. The second rulebook is the Canada-United States-Mexico Agreement. It governs much of the same conduct that Bill C-36 regulates, the movement of data across the border, the treatment of foreign service suppliers, and the disclosure of how software works. It is not enforced in a Canadian court, and the penalty for getting it wrong is not that the law is read down or struck. The penalty is retaliation against unrelated Canadian exports. A privacy rule written in Ottawa can be paid for by a mill in British Columbia or a farm in Quebec.

This is the gap that closed on the Digital Services Tax. That measure was not legally fragile. It was institutionally undefended. The United States had already taken the tax to formal consultations under CUSMA Chapter 31 in August 2024, and when Washington suspended trade talks the next summer, Canada folded within days. There was no advisory architecture, no prepared trade position, and no appetite to hold the line when the pressure came. Bill C-36 is a genuine privacy bill, not a revenue grab, and it deserves better than the tax revenue it receives. But several of its provisions carry the same structural flaw: they were designed without a reading of the treaty that the United States will read very closely.

Where the bill meets the treaty

Start with the provision that matters most. Section 57 requires that, before disclosing or transferring personal information outside Canada, an organization must conduct a privacy impact assessment and implement measures to mitigate the risks identified in the assessment.11 Read in isolation, that sounds prudent. Read against CUSMA Article 19.11, it is exposed.12 Article 19.11 prohibits measures that restrict the cross-border transfer of information by electronic means. A mandatory assessment before every outbound transfer is a restriction on the transfer. The treaty allows such a measure only through a narrow valve in Article 19.11(2): the measure must be necessary to achieve a legitimate public policy objective, must not discriminate arbitrarily, and must not be more trade restrictive than necessary.

Privacy clears the first condition. The third condition is the one in which section 57 is most vulnerable. Canada would require a privacy impact assessment before every transfer out of the country. The European Union, which operates the strictest privacy regime in the world, does not demand a fresh assessment for each routine transfer; it relies on standing transfer mechanisms instead.13 A Canadian rule that is more demanding than the European one is hard to defend as no more trade restrictive than necessary. That is the most powerful argument the United States has, and section 57 hands it over.

Then there is the self-inflicted wound. The government’s own backgrounder describes section 57 as designed to “enhance Canada’s digital sovereignty by requiring robust privacy safeguards and risk assessments before personal information is transferred outside Canada.”14 I have spent years arguing that Canada needs digital sovereignty. I did not expect the word to be stapled to the one instrument the treaty most directly forbids. The chapeau of Article 19.11(2) excludes any measure applied as a disguised restriction on trade. A government that publicly labels a data-flow restriction as a sovereignty measure has drafted the opening paragraph of the American complaint. Intent matters in treaty disputes, and the press release is now part of the record.

The exposure does not stop at section 57. Pair it with section 72, the data mobility framework. If mobility rights are built into regulation so that data moves easily between Canadian organizations, while every transfer abroad must clear an assessment, the practical result is that information is more portable within Canada than across the border. That is not localization by command. It is localization by architecture, and Article 19.12, which forbids data localization requirements, contains no exception of any kind.

The bill’s treatment of automated decisions raises a third problem. Bill C-36 defines an automated decision system to include machine learning, deep learning, and neural networks, and it requires organizations to be transparent about how those systems make significant decisions about individuals.15 CUSMA Article 19.16 prohibits requiring access to the source code or algorithms of software as a condition of doing business. Explaining how a neural network reached a decision can shade into disclosing how the model works. The safe path is the one the treaty already marks out: build the transparency duty as an individual right, the right of a person to know why a specific decision was made about them, rather than a standing demand that firms reveal the logic of their systems to the regulator on request. The difference lies between a defensible measure and a trade dispute.

Two further provisions sit lower on the list but belong on it. The surveillance pricing prohibition and the global revenue penalty are written in neutral language, yet in practice, both fall almost entirely on large American platforms, because those are the firms that price based on behavioral data and earn the global revenue the penalty is scaled to reach. That asymmetry invites a non-discrimination claim under Article 19.4, even where the text names no nationality. The personal data shelter in Article 19.8 provides real cover for the bill’s privacy core, but that shelter carries its own chapeau, and it does not protect a measure that operates as disguised discrimination.

Taken one at a time, most of these provisions can be defended. Taken together, they read as something the United States already has a name for. The 2026 National Trade Estimate report singles out the Online Streaming Act and Quebec’s French-language discoverability law for monitoring, noting, in each case, their USMCA implications by name, and lists the Online News Act alongside them.16 Bill C-36 risks becoming the next entry on that list. In the year of the Joint Review, another entry is not a footnote. It is a target.

The contradiction that is not one

Here is where the reader who follows both halves of my work will raise a fair objection. For years, I have argued that Canada must keep its data under Canadian jurisdiction and legislate the code that governs Canadian life. Now I appear to be arguing that Canada cannot restrict the flow of its data abroad. Which is it?

It is both, and the two resolve through sequence. Sovereignty is built at the infrastructure layer, not the regulatory layer. Bill C-36 tries to achieve sovereignty by regulating the data as it leaves. That is the weakest method available, because it is the one CUSMA most directly prohibits, and the one Washington most easily punishes. Real sovereignty is achieved by building the capacity that gives the data somewhere to stay.

Sovereignty by construction is compatible with the treaty. Sovereignty by restriction is a breach waiting for a respondent.

Switzerland understood this. When the Swiss launched Apertus, a fully open and sovereign large language model trained on their own supercomputer and hosted on their own cloud, they did not pass a law forbidding Swiss data from leaving the country.17 They built a home for it. Sovereignty by construction is compatible with the treaty because building your own infrastructure imposes no obligation on anyone. Sovereignty by restriction is a breach waiting to happen because it works by denying others access to your market.

Canada has the raw materials. The federal government has committed roughly two billion dollars to a sovereign AI compute strategy, with hundreds of millions for domestic supercomputing.18 What Canada has not done is connect that construction to its regulation. Bill C-36 legislates a wall and funds a foundation, and never notices that the wall would not be necessary if the foundation were finished.

The deeper rule is one I have stated before and will state again. You can only restrict what you can replace. Canada has not built the replacement, so the restriction is empty, and worse than empty, because it invites retaliation without delivering autonomy.

A privacy impact assessment filed before every transfer does not build a single server on Canadian soil.

A privacy impact assessment filed before every transfer does not mean a single server is built on Canadian soil. It does not bring home one byte of Canadian data. It raises a paper wall that the treaty will eventually pull down, and that the United States can impose tariffs on while it stands.

The same political will, spent on sovereign cloud and sovereign compute, would produce sovereignty that survives the Joint Review instead of sovereignty that provokes it.

The Prime Minister named the right lesson in Ireland: build out and diversify. Bill C-36 reaches for the other hand.

How to make the bill stronger

This is not a case for abandoning Bill C-36. It is a case for finishing it properly, and the timing is favorable. The bill has had only its first reading. Its hardest edges live in regulations that have not been written. Parliament rises for the summer on June 19 and returns on September 21.19 The months in between are the design window, when this should be fixed.

Three corrections would carry most of the weight. First, recalibrate section 57. Replace the requirement for an assessment before every transfer with a risk-based standard keyed to the data’s sensitivity and the safeguards at the destination, applied evenhandedly regardless of where the data is going. That protects Canadians and survives Article 19.11(2), and it is a better privacy policy because an assessment is required for every transfer, regardless of risk, making it a compliance ritual rather than a protection.

Second, frame the automated decision duty as an individual right to an explanation of a specific decision, not as a standing disclosure obligation owed to the regulator. The first is defensible under the treaty. The second walks into Article 19.16.

Third, and most important, pair every act of restriction with an act of construction. If the government wants Canadian data to stay in Canada, the durable lever is not a transfer assessment. It is a Canadian place for data to be sovereign cloud, sovereign compute, and the unified digital authority that Canada still lacks, but that Switzerland has built. Restriction without construction is the pattern that led to the loss of the Digital Services Tax. Construction is what makes restriction unnecessary.

France is about to test whether it can hold a digital measure that Canada could not. France sits outside CUSMA. Canada sits inside it, which means Canada has less room, not more. The honest reading of this week is that the digital file will sit at the center of the Joint Review, that Washington will use both tariffs and access as instruments, and that Canada will be measured by what it has built, not by what it has banned.

Bill C-36 can be the start of that building. It can protect Canadians, satisfy the treaty, and advance real sovereignty at the same time, but only if the government understands which hand to reach with. Sovereignty is not granted by a clause, and it is not secured by a ban. It is coded, built, and defended. Switzerland built a model. So far, Canada has drafted an assessment form.

Prof. Barry Appleton is Managing Partner of Appleton & Associates International Lawyers LP, Co-Director and Distinguished Senior Fellow at the Center for International Law at New York Law School, and Interim Director of the Balsillie Legal Advisory Centre at the Balsillie School of International Affairs. He writes Appleton’s Clause & Effect on international trade law, digital sovereignty, and Canadian economic statecraft.

This post is public. If you found it useful, please share it with a colleague.

© 2026 Barry Appleton. All rights reserved.

Notes