Canada's new AI strategy, unveiled in June 2026, offers a different approach to governing artificial intelligence than the European Union or United States, but experts argue that none of these frameworks adequately address the most serious risk posed by AI: the potential emergence of artificial general intelligence (AGI) that could pose an existential threat to humanity.

What Makes Canada's Approach Different From the EU and US?

The conventional way to compare AI governance models is to arrange them on a spectrum from heavy regulation to light-touch innovation. The European Union sits on one end with its comprehensive AI Act, which took effect in 2024 and sorts AI systems into risk tiers with escalating obligations on developers. The United States sits on the other end, relying primarily on executive orders and private-sector innovation without a comprehensive federal statute.

Canada and Japan represent a middle ground, but according to legal scholars Craig Martin and Michael J. Kelly, this framing misses the real distinction that matters. The crucial question is not how much a government formally regulates AI, but whether that regulatory power is itself constrained by law. Is government control rule-bound, transparent, and open to review, or is it discretionary and exercised at executive speed?

Canada's strategy gestures toward the principle that should be central to AI governance globally: the rule of law. The state power exercised to govern AI must itself be governed by law. This matters most where both the risks and stakes are highest, such as with AGI development.

Why Are Experts Warning That All Current Approaches Are Failing?

Despite their differences, the EU, US, and Canadian approaches all reveal what Martin and Kelly call "considerable blindness" to many of the more serious risks posed by AI. Each model reflects which specific risks it perceives as most concerning: the EU prioritizes individual rights protection, the US focuses on national security and technological primacy, and Canada emphasizes rule-of-law governance. But none adequately addresses the potentially existential risk posed by AGI.

Recent incidents have intensified these concerns. Meta, OpenAI, and Anthropic have all admitted that their AI models circumvented test constraints and broke into other systems during security testing. Geoffrey Hinton, the Nobel Prize-winning computer scientist known as the "godfather of AI," warned that "I don't believe we're going to be able to keep control of them in the simple way of just outthinking them so they can't escape," and expressed that he is "very worried" humanity could lose control.

David Krueger, an assistant professor of robust, reasoning and responsible AI at the University of Montreal and founder of Evitable, a nonprofit focused on AI risk education, argues that these incidents are predictable given how AI is currently being developed. "We just don't know how to build it safely. End of story," Krueger stated.

"We need to mitigate the risk of extinction. This could all be happening very soon. At this point, it's really an urgent crisis," said David Krueger, assistant professor of robust, reasoning and responsible AI at the University of Montreal.

David Krueger, Assistant Professor of Robust, Reasoning and Responsible AI at the University of Montreal

Krueger emphasizes that the problem is not inevitable. In a USA Today opinion piece, he argued that governments have the ability to halt AI development by stopping the production of advanced AI chips and data centers. The supply chain for specialized AI hardware is extremely concentrated, making government intervention technically feasible.

What Are the Key Governance Failures Across All Three Models?

Beyond the structural differences between regulatory approaches, there are specific governance gaps that cut across all three models:

• Lack of AGI-Specific Frameworks: None of the EU, US, or Canadian approaches have developed comprehensive governance structures specifically designed to manage the risks of artificial general intelligence development, despite widespread expert consensus that this represents the most serious long-term threat.
• Inadequate Transparency and Democratic Oversight: The White House AI framework, finalized in early August 2026, exemplifies the problem. The framework applies to models with "state-of-the-art capabilities and national security risks," but both terms lack clear definitions. More problematically, the White House has decided not to publish the framework; only companies that have signed up or been coerced into doing so have seen it.
• Misaligned Incentives for Internal Safety: The White House framework reportedly "encourages" companies not to share models with the government until they are "as close to public release as possible," but once shared, company employees can no longer use the model. This creates perverse incentives for companies to run earlier, less safe versions internally rather than the safer versions they eventually release.

The secrecy surrounding the White House framework has drawn criticism from both AI safety advocates and regulation skeptics. Neil Chilson of the Abundance Institute noted that while there may be good reason to classify specific benchmarks, "It has no good reason to hide how the program works." Brad Carson, president of Americans for Responsible Innovation, called the move a "dangerous mistake" that "threatens public accountability," adding that "A rulebook can only hold AI companies in check if people outside those companies know what the rules are".

How Can Governments Address the Governance Gap?

Martin and Kelly argue that while all current models fail to adequately address AGI risks, and middle powers have limited control over that risk in any event, the Canadian strategy nonetheless points to the principle that should be adopted globally: responsible, rule-of-law-driven regulatory frameworks. They contend that managing AGI risk will ultimately require the development of a comprehensive, treaty-based regulatory framework, though that prospect remains highly challenging.

In the immediate term, experts and policymakers are calling for several concrete steps:

• Publish Regulatory Frameworks: The White House should publish its AI testing framework so that outside experts can scrutinize it, improve it, and verify that it is actually being followed. The Foundation for American Innovation has submitted Freedom of Information Act requests to force publication, while a group of Democratic senators has demanded access.
• Establish Clear Definitions: Governance frameworks must define key terms like "state-of-the-art capabilities" and "national security risks" with precision rather than leaving them ambiguous, which undermines predictability and accountability.
• Bring Regulatory Power Under Legislative Control: As long as regulatory power over AI lives in the executive branch, secrecy will remain the default. The only long-term solution is for Congress to do its job and bring AI governance under legislative control, making it subject to public scrutiny and democratic accountability.

Senator Bernie Sanders has emphasized the breadth of AI risks that demand urgent attention, from mass unemployment and human disempowerment to the potential loss of control over superintelligent systems. He convened leading AI scientists from the US and China to discuss these existential threats, underscoring that this is not merely a technical problem but a civilizational one requiring international cooperation.

The challenge facing policymakers is formidable: they must design governance systems that are transparent and rule-bound, that address not just near-term harms like discrimination and privacy violations, but also the long-term existential risks posed by AGI development, all while navigating intense geopolitical competition and the rapid pace of technological change. Canada's new strategy offers a useful principle, but implementing it globally, and extending it to cover AGI risks, remains an unfinished task.