"AI for All": What is the Value of a Voluntary Code without a Rebuttable Presumption of Conformity

Create a watch on this topic Receive upcoming developments if any, with an AI watch configured from this article.

Create my free watch

According to the Canadian Centre for Policy Alternatives (CCPA), Canada is the only G7 country without a specific federal binding regulatory framework for AI. It is in this vacuum that Mark Carney launched the national strategy "AI for All" on June 4, 2026, at the Toronto General Hospital, articulated around six pillars and accompanied by a budget exceeding 2.3 billion CAD. Since the parliamentary extension on January 5, 2025, the AIDA - the Artificial Intelligence and Data Act under Bill C-27 - has died, and no binding federal AI law has replaced it. The CCPA and The Walrus describe the device as largely voluntary, due to the absence of binding obligations.

A single G7 without a binding federal framework

According to the Canadian Centre for Policy Alternatives, Canada is the only G7 country without a specific federal binding regulatory framework for AI. The finding deserves nuance: Ottawa and the provinces are already operating through partial levers - the Treasury Board Directive on Automated Decision-Making (2019), the LPRPDE, Quebec's Bill 25, and the OSFI guidelines for the financial sector. These instruments cover segments (federal administration, private life, finance) without forming a unified AI framework, and the CCPA argues that they are insufficient to regulate high-risk systems deployed in the private sector. The most immediate constraint for a Canadian actor exposed to the European market comes from elsewhere: the AI Act applies to providers and operators whose AI systems are placed on the market or used within the Union, regardless of the country of establishment. A Canadian company deploying a high-risk system on the European market is therefore subject to the regulation, regardless of the status of federal law in Ottawa.

Voluntary does not mean without effect – it still requires anchoring

The distinction from the European experience does not lie in the nature of the instrument: the Code of Good Practices for General Purpose AI Models published by the European Commission is also a voluntary instrument. Its normative effect comes from its integration with the AI Act, which includes a provision stating that signatories of the GPAI Code benefit from an opposable presumption of compliance with obligations applicable to general-purpose model providers. The signature produces a tangible legal benefit; non-signature exposes one to the need to demonstrate compliance through other, more burdensome means. Without a similarly binding federal framework on the Canadian side, the "AI for All" strategy places the normative pressure on reputation and market impact – a mechanism that may produce alignment in concentrated and exposed sectors, as illustrated by the UK's pre-Online Safety Act voluntary codes, but which does not create the same level of enforceability. The Walrus reports that a survey cited in its analysis places the share of Canadians willing to trust AI at 34%, compared to 75% expecting state regulation: the reputational pressure anticipated by Ottawa is taking place in an environment where the demand for a legal framework is explicit.

An ecosystem backed by three institutes, an infrastructure described as « embryonic »

The government counts more than 3,500 Canadian companies developing AI solutions, collectively raising over 37 billion CAD in venture capital. The macroeconomic impact projections advanced by Ottawa belong to another category: Ottawa states it aims for the creation of 250,000 jobs and a 3% increase in GDP - figures presented as government objectives, not as third-party audited forecasts. The official document also explicitly acknowledges that Canada's sovereign computing capacity « remains embryonic, particularly in cloud computing, and will require significant investments to reduce reliance on foreign providers » - an admission that, within the text itself, limits the scope of the sovereignty pillar.