Canada: the bill that died and the directive that lives

Lesson 2 of 5 in The UK, Canada, Brazil — and the Missing Map: Africa, the Middle East, Latin America.

Canada nearly became the first Western democracy with a horizontal AI statute — and ended up with none. The Artificial Intelligence and Data Act (AIDA) was Part 3 of Bill C-27, the Digital Charter Implementation Act introduced in June 2022. It would have imposed duties on ‘high-impact’ AI systems: risk assessment, mitigation, monitoring, transparency, and criminal penalties for reckless deployment.

Then it spent two and a half years in committee being taken apart. Critics attacked the ‘high-impact’ definition — left almost entirely to future regulations, so Parliament was being asked to pass a law whose scope nobody could state. They attacked the institutional design: ISED, the innovation ministry, would both champion the AI industry and enforce the law against it, with the ‘AI and Data Commissioner’ a ministry official rather than an independent regulator. Amendments piled up; the government produced a companion document rewriting the bill’s intent mid-passage; civil society and industry both testified against it, for opposite reasons.

On 6 January 2025, Parliament was prorogued. Every bill on the order paper died — including C-27. AIDA did not fail a vote; it simply ran out of parliament.

Bill C-27’s journey — and the trapdoor at the end

  1. Introduced (June 2022)

    Bill C-27 bundles privacy reform (CPPA), a tribunal, and AIDA into one bill — a packaging choice that slowed all three.

  2. Second reading passed

    April 2023 — the last stage C-27 ever completed in the House.

  3. INDU committee study

    Eighteen months of hearings, 100+ witnesses, government amendments rewriting core provisions mid-study.

  4. Committee stalls

    Amendment gridlock plus wider parliamentary dysfunction through 2024.

  5. Prorogation — 6 January 2025

    Parliament prorogued; all pending bills die regardless of merit or progress.

  6. AIDA dead — no AI statute

    Any successor must restart from first reading in a new Parliament.

What governs Canadian AI now? A voluntary code and a binding directive — one for industry, one for government.

The Voluntary Code of Conduct on Advanced Generative AI (September 2023) collects safety, fairness, transparency and monitoring commitments from signatory firms — Canada’s interim answer while legislation restarts. The machinery around it kept moving even as the law did not: Canada signed the CoE Framework Convention on 11 February 2025, stood up the Canadian AI Safety Institute in November 2024, and in May 2025 appointed its first Minister of AI and Digital Innovation (Evan Solomon), with a task force renewing the national AI strategy. Canada funds AI like a superpower — the Pan-Canadian AI Strategy (2017) was the world’s first national AI strategy — and regulates it, federally, almost not at all. PIPEDA, the private-sector privacy law C-27 would have replaced, remains unreformed.

The exception — and it is a genuinely influential one — is the Treasury Board Directive on Automated Decision-Making (2019, periodically amended). It binds federal government use of automated decision systems, and it works through an instrument you will meet everywhere: the Algorithmic Impact Assessment (AIA), a public, standardised questionnaire (about 80 questions on data, impact, transparency and recourse) that scores a system into one of four impact levels. The level then dictates the obligations — a graduated, published, auditable scheme that jurisdictions from Brazil to the EU studied closely.

Canada’s Directive on ADM — obligations scale with the AIA impact level
Impact levelTypical decisionsWhat is required

Level I — little impact

Reversible, brief effects

Plain-language notice that automation is used; basic explanation available

Level II — moderate

Reversible, short-term effects on rights or interests

Level I duties plus peer review or published specifications and human intervention points during design

Level III — high

Hard-to-reverse or ongoing effects (e.g. benefits eligibility)

Qualified peer review, meaningful explanations, human-in-the-loop for adverse decisions, recourse channels, monitoring

Level IV — very high

Irreversible or perpetual effects on individuals or communities

The full stack: external peer review, approval gates at the deputy-head level, the strongest oversight and explanation duties

Key terms: algorithmic impact assessment, human-in-the-loop, Council of Europe Framework Convention on AI, AI safety institute

Interactive checkpoint quiz (2 questions) — open this page in a browser to take it.