On April 28, 2026, the Liberal government used its Spring Economic Update to unveil the six pillars of the long-delayed national AI strategy it is branding 'AI for All.' The framework — protecting Canadians and democracy, empowering citizens, powering adoption, building sovereign AI infrastructure, scaling Canadian champions, and building global alliances — is Ottawa's chosen successor to the Artificial Intelligence and Data Act (AIDA), the Bill C-27 chapter that died on prorogation on January 6, 2025 after nearly three years of parliamentary stalemate.
AI Minister Evan Solomon has been explicit that the successor will not be a resurrected AIDA. In an interview earlier this spring he said Canada needs regulation that is 'light, tight, and right', and that the government will not return to a single omnibus AI statute. Instead, the AI-specific obligations are being broken up across a modernized privacy law (a PIPEDA refresh including new data-sovereignty and IP protections), a promised online harms bill, and procurement-side conditions for the federal government's own AI use.
The strongest case for an omnibus AI act
It is worth steelmanning the opposite view before arguing against it. AIDA's defenders had a serious point: the EU has its AI Act, the UK has its sectoral AI rules backed by central coordination, and the US has executive-branch directives plus a thicket of state laws. A coherent Canadian statute, with a dedicated commissioner and risk-tiered obligations, would have given developers a single forum, given citizens a single point of redress, and given trading partners legible interoperability. Without that, fragmented enforcement under a half-dozen agencies risks looking improvised — and lawyers, not engineers, end up mapping where any given model sits in the regulatory grid.
That case is real. The problem is that AIDA, as drafted, was not the law that would have delivered it. McInnes Cooper's post-mortem documents the structural defects: a definition of 'high-impact' AI broad enough to sweep in general-purpose models by default, an AI Commissioner who would have been a civil servant reporting directly to the Industry Minister rather than an independent regulator, a near-total carve-out for federal government use of AI, and a constitutional vulnerability under the Supreme Court's 2023 Reference re Impact Assessment Act ruling on the limits of federal jurisdiction over provincial activity. The bill spent more than 30 months in committee and never reached third reading. Prorogation killed it, but it was failing on its own.
What 'light, tight, and right' looks like in practice
The Spring Economic Update — the formal text is in Chapter 1 of Budget Canada's update document — frames Pillar 1 ('Protecting Canadians and safeguarding democracy') almost entirely in terms of privacy and online safety law rather than horizontal AI rules. That is a meaningful shift. The PIPEDA refresh will address copyright, intellectual property, and data sovereignty in the context of AI training and decisioning. The promised online harms bill picks up the synthetic-media and deepfake concerns that AIDA had tried to bundle inside an industrial-regulation chassis. National security and election integrity get their own line items rather than being delegated to a single AI commissioner.
This disaggregated approach has two underappreciated advantages. First, it constrains regulators to act through statutes that already have judicial interpretation and constitutional footing — PIPEDA has been litigated for two decades. Second, it preserves space for the courts to do the slow, fact-specific work of mapping AI onto existing doctrine. Solomon has explicitly said Ottawa wants to let the live AI-copyright cases 'play out' before locking remedies into legislation. That is the right instinct: prematurely codified rules in this area would freeze a market still figuring out how compensation, licensing, and fair-dealing defaults should work.
The industrial half of the strategy is where the real money is
Pillars 3–5 — adoption, sovereign infrastructure, and scaling Canadian champions — are not regulatory pillars at all. They are an industrial-policy program riding on the $2-billion, five-year Canadian Sovereign AI Compute Strategy that Ottawa announced in December 2024 under Budget 2024 authority. That envelope breaks into roughly $1 billion for public supercomputing infrastructure, $700 million for private–public compute partnerships (the AI Compute Challenge), and $300 million for an AI Compute Access Fund to subsidize researcher and start-up access. Applications opened in June 2025.
This is the part of the strategy most likely to matter for Canadian AI startups in the next 24 months. The growth-capital and government-procurement levers in Pillar 5 will succeed or fail on whether Treasury Board can credibly use Ottawa as an anchor customer. The 11,000-plus public submissions and 28-member expert task force that fed into 'AI for All' produced a clearer mandate on this point than on regulation: Canadian capital is too patient to scale homegrown AI champions on its own, and a domestic compute base is the precondition for keeping IP, headquarters, and talent in country.
The risk: under-built rather than over-built
The legitimate concern with the unbundled approach is enforcement gaps. If PIPEDA reform slips, the online harms bill stalls in committee (as its predecessors did), and a dedicated AI watchdog never gets stood up, Canada could end up with neither AIDA's heavy hand nor a functioning lighter-touch alternative. The strategy's credibility now turns on parliamentary execution across three or four discrete bills — a harder political ask than passing one omnibus.
But that is the right risk to be running. The wrong risk — locking in a poorly drafted definition of 'high-impact' AI for a decade, with a non-independent commissioner and a federal-government carve-out — is the one AIDA represented. 'Light, tight, and right' is a slogan; whether Solomon delivers on it depends entirely on how the privacy bill, the harms bill, and the compute money are scoped over the next eighteen months. The Spring Economic Update is a credible starting point, not a finish line.