On June 4, 2026, Prime Minister Mark Carney unveiled AI for All, Canada's refreshed national AI strategy, promising $2 billion in federal investment, 250,000 new jobs, and AI adoption rising from 12% of Canadian businesses today to 60% by 2034. Six days later, the government tabled Bill C-34, the Safe Social Media Act, adding binding crisis-intervention and transparency duties for AI chatbot operators. Together, the two measures mark Ottawa's most consequential AI policy move since the Artificial Intelligence and Data Act (AIDA) died on Parliament's order paper when Justin Trudeau's government fell and Parliament was prorogued in January 2025.
AIDA's Failure Was Not an Accident
Introduced as part of Bill C-27 in June 2022, AIDA spent more than two years stalled in committee, attracting objections from across the spectrum. Industry argued its high-impact AI system definitions were too vague to comply with. Labour groups and Indigenous nations warned it lacked rights-based protections. Privacy advocates faulted the absence of independent regulatory oversight. The Montreal AI Ethics Institute's post-mortem was direct—the act was "vaguely described and insufficient for protecting Canada against harmful AI impacts," developed largely behind closed doors with a selective group of stakeholders. When Parliament was prorogued, AIDA did not die of political misfortune alone: it had spent years failing to build a coalition capable of surviving one.
That history forms the strongest argument for the government's current approach. Opponents of AI for All are right that a voluntary certification program and sector-specific literacy funding are not a substitute for binding legal obligations on AI developers—Canadian citizens in healthcare decisions or credit assessments deserve legal recourse, not industry goodwill. But those critics must also grapple with AIDA's lesson: what does a mandatory comprehensive framework accomplish when three years of drafting produced something too vague to enforce and too blunt to innovate under?
What the Strategy Actually Contains
AI for All is best understood as a demand-side growth strategy grafted onto targeted supply-side obligations. The $2 billion envelope funds a national AI supercomputer, sovereign compute infrastructure, and a Canada Trusted AI Certification program built on voluntary standards from the Standards Council of Canada. The $500 million LIFT program offers incentives to small and medium enterprises for AI adoption. A National AI Literacy Initiative targets one million post-secondary students; 90,000 work placements aim to build the talent pipeline across six priority sectors.
The regulatory layer is deliberately narrow. Rather than a single comprehensive AI Act, Ottawa has chosen to modernize existing legislation—privacy law, election integrity rules, deepfake provisions—and pair those reforms with sectoral legislation. Bill C-34 is the first installment of that sectoral approach.
Bill C-34: Targeted, Functional, Incomplete
The Safe Social Media Act bundles a Digital Safety Commission, a social media access regime for under-16s, platform duty-of-care requirements, and AI chatbot-specific obligations into a single bill. Legal commentator Michael Geist has argued this "everything-all-at-once approach" risks long implementation delays because it simultaneously invites opposition from every direction.
On the chatbot side specifically, the obligations are concrete and proportionate. Section 51 of the bill requires operators to interrupt interactions and redirect users to crisis services with human access whenever a user expresses suicidal ideation or an intention to harm themselves or others. Deception provisions bar chatbots from impersonating licensed professionals or deploying manipulative engagement techniques. Digital safety plans must be publicly filed with the Digital Safety Commission, and accredited researchers gain structured access to operator safety data.
These duties target documented harms—cases where AI chatbots have reinforced self-harm narratives, simulated therapeutic relationships, and provided unaccountable pseudo-professional advice—without mandating how systems must be built. Crisis-intervention duties do not restrict speech; they ensure a safety exit exists when users need it. That is the right calibration for a free-speech-respecting regime.
What Bill C-34 does not do is apply risk-tiered governance to the broader AI stack. A general-purpose model deployed in hiring, credit scoring, or diagnostic support remains entirely outside the bill's scope. There is no obligation to register high-impact systems, disclose training data, or conduct algorithmic impact assessments.
The Governance Gap Is Real
The government's position is that consequential AI decisions will be addressed through promised privacy law modernization and future sectoral regulation. That is plausible in theory. In practice, Canadian businesses and citizens have watched AI regulation promised and deferred since at least 2017. AI systems are already deployed across healthcare, policing, and financial services in Canada, often with no consistent public accountability mechanism.
Legal and policy critics are right that the distance between "we will modernize privacy law" and actual binding obligations on AI deployers is where citizens are most exposed. A voluntary certification path creates reputational incentives for compliance but no legal recourse when those incentives fail. Each legislative cycle that passes without binding rules for high-impact AI is not merely a gap—it is an accumulation of unaccountable deployments that become harder to unwind as they normalize.
A Defensible Bet—With a Hard Condition
The case for the current approach is not that it is ideal, but that it is executable. AIDA proved that a maximalist comprehensive bill generates maximalist opposition and delivers nothing. A strategy that builds AI capacity, attaches binding duties to acute harms—chatbot crisis intervention is a genuine advance—and commits to layering further obligations through modernized sectoral law is internally coherent, provided the follow-through materializes.
The real test of AI for All will not arrive at its June 2026 launch. It will arrive when Ottawa tables its revised privacy legislation, when the Digital Safety Commission issues its first enforcement actions under C-34, and when the government must answer whether the promised high-impact AI protections were always planned or were always placeholders. Canadians are currently underwriting a governance bet: that the government which buried AIDA's flawed comprehensiveness will replace it with something binding, not with perpetual deferral dressed as flexibility.
The bet is not obviously wrong. The odds are not obviously good.