On May 4, 2026, AI Minister Evan Solomon told reporters that Canada's long-delayed national AI strategy will now explicitly track artificial intelligence's effect on jobs and the labour market. Days earlier, the April 29 spring economic statement previewed the strategy's six pillars — led by "protecting Canadians and safeguarding our democracy" through new privacy and online-safety laws, alongside sovereign compute, AI skills, adoption support, scaling domestic champions, and global alliances. Together they mark a clear tonal shift from the adoption-first, stop-"over-indexing on warnings and regulation" posture Solomon struck when he took the file a year ago.
A third act on AI governance
This is Ottawa's third move on AI governance in under four years, and the shadow of the second hangs over it. The Artificial Intelligence and Data Act (AIDA), bundled into Bill C-27, died on the order paper in January 2025 when Parliament was prorogued. Rather than reintroduce a standalone AI statute, the current government has said it will regulate "the design, development and deployment of AI technologies through privacy legislation, policy and investment." The six-pillar framework is the policy-and-investment half of that bet — and, notably, most of it is not regulation at all.
The case for the pivot is real
It would be a mistake to dismiss the shift as mere caution. The strongest argument for foregrounding safety and labour is that legitimacy is a precondition for adoption, not a tax on it. Ottawa's 2025 consultation drew more than 11,300 respondents — the largest public consultation in the history of Innovation, Science and Economic Development Canada — and job displacement, transparent oversight, and risk-based rules surfaced as dominant themes. A government that ignored those anxieties would forfeit the public trust that diffusion of the technology depends on. McGill's Sonja Solomun called the safety-first framing "a discernible shift," and she is right that durable AI deployment rests on it.
Granted. But a strategy is judged by what it builds and what it binds, and on both counts the pillars cut in a pro-innovation direction more than the safety-first framing suggests.
Capacity, not rules, is the real win
The most valuable items on the list are not new legal duties — they are capacity. The $2-billion Canadian Sovereign AI Compute Strategy, announced in Budget 2024 over five years, splits into up to $1 billion for public supercomputing, $700 million for a private-sector AI Compute Challenge, and $300 million in an AI Compute Access Fund. Its flagship, the $890-million Sovereign Compute Infrastructure Program, opened to applicants on April 15, 2026. Compute, skills training, and SME adoption support move Canada's frontier outward; they create the firms and the workforce that any future rules will govern. A government that ships these and legislates lightly will have done more for both safety and prosperity than one that inverts the order.
AIDA's real lesson was vagueness, not ambition
The risk is that the regulatory pillar repeats AIDA's central error. AIDA failed not because it tried to govern AI, but because its core terms — what counted as a "high-impact system," what obligations attached — were punted to regulations that did not yet exist, giving businesses nothing concrete to plan around. The forthcoming privacy and online-safety bills should learn from that. If they pair undefined scope with new labour-impact mandates of uncertain shape, Canada risks the worst of both worlds: compliance cost without compliance clarity, the precise combination that stalls the small and mid-sized adopters the strategy claims to champion.
The online-safety leg warrants the closest scrutiny. The Electronic Frontier Foundation has warned that Bill C-22 is a repackaged version of last year's surveillance proposals. Folding AI chatbots into an online-harms regime built on broad monitoring and takedown duties — as Solomon has floated — would import those free-expression and surveillance problems straight into AI policy. Speech-adjacent AI tools deserve proportionate, harm-specific rules, not a general-purpose content-policing apparatus retrofitted to cover models.
Track, don't pre-empt
Tracking AI's labour-market impact is sound as measurement. Good data on displacement and reskilling needs is exactly what a $2-billion industrial bet should generate. It becomes a problem only if "tracking" hardens into pre-emptive limits on where and how firms may deploy — restrictions that would push the very investment Canada is trying to attract toward jurisdictions that merely monitor outcomes.
The proportionate path is visible in the pillars themselves. Ship the compute, skills, and adoption commitments now; they need no new statute. Legislate narrowly and concretely on privacy, with defined scope and enforceable, technology-neutral obligations. Resist bolting AI governance onto a contested online-harms bill. Canada's pivot toward safety and jobs is defensible — even overdue on trust grounds. Whether it helps or hurts depends entirely on whether the eventual bills are sharper than the act that died, or just vaguer with a friendlier label.