Quebec's AMF AI Guideline: what regulated firms need to know

The AMF AI Guideline is a guideline published by the Autorite des marches financiers, Quebec's financial regulator, on the use of artificial intelligence by the financial institutions it supervises. It is one of the first dedicated AI guidelines from a Canadian financial supervisor, and it is principles-based rather than a checklist. The principles are the ones now recurring across Canadian supervisors: AI systems should be transparent and explainable, fair and free of unjustified bias, robust and tested, and accountable to a human. In practice that means knowing which AI systems you run, documenting what they do, validating them, monitoring them, and keeping a person responsible for the outcomes. The obligations come into force on May 1, 2027, which gives firms in scope a runway to build the governance rather than scramble at the deadline. Its significance reaches beyond Quebec, because it signals where Canadian supervisory expectations are heading.

The obligations come into force on May 1, 2027. That date is the point, not a distant horizon. Building AI governance is a program rather than a switch, so the institutions in scope are expected to use the runway between now and the deadline to put the framework in place rather than scramble at the last moment. Standing up a model inventory, documenting what each AI system does, validating and monitoring it, and assigning human accountability all take time to design and embed, which is precisely why the guideline gives a runway rather than immediate effect. A firm that treats May 1, 2027 as the day to start will be building under pressure; a firm that starts now can build calmly and test the framework before it is examined against it. The runway is the opportunity, and the practical reading of the effective date is to use it, not wait for it.

The guideline applies to the financial institutions the AMF supervises in Quebec, including insurers, deposit institutions, financial services cooperatives, and trust companies. Those are the firms directly bound by it, and for them the obligations come into force on May 1, 2027. Firms outside the AMF's scope, including most reporting entities and many that sit outside Quebec entirely, are not bound by the guideline. But it still matters to them, because when a Canadian financial supervisor publishes formal AI expectations, it sets a reference point that other regulators, banking partners, and auditors read. The principles it sets out, transparency, fairness, robustness, and accountability, are the same expectations surfacing across Canadian supervisors, so the guideline reads as a direction of travel rather than a Quebec-only exception. A firm not directly covered is still well advised to treat it as a preview of where the bar for AI in financial services is moving.

Indirectly but meaningfully. The AMF principles, transparency and explainability, fairness, robustness, and accountability, are largely the same controls a defensible AML program already needs for its automated decisions, so a firm with strong AML governance is most of the way to meeting them. If your automated screening, scoring, and verification decisions are already documented, explainable, validated, and owned by a named person, you have most of an AI governance framework in hand. The remaining work is to formalise that discipline into the structured framework a supervisor expects to see, rather than build it from nothing. The overlap runs the other way too: the AMF principles echo OSFI's model risk expectations and the wider federal direction of travel, so an AML program built to explain and govern its automated decisions is aligning with several supervisory reference points at once. For an AML function, the guideline is less a new burden than a confirmation of the direction already underway.

No, not directly. The guideline binds only the financial institutions the AMF supervises in Quebec, so a firm outside that scope, or outside the province, has no obligation under it. But it is one of the first dedicated AI guidelines from a Canadian financial supervisor, which is exactly why it matters beyond Quebec. When a regulator publishes formal AI expectations, it sets a reference point that other regulators, banking partners, and auditors read, and the convergence around transparency, fairness, robustness, and accountability is clear enough that AI governance has become a Canadian financial-services expectation, not a Quebec one. A firm that waits for its own regulator to mandate AI governance is choosing to build under pressure later instead of calmly now. The practical reading for a firm outside Quebec is to treat the guideline as a signal of where the bar is moving, and to build the governance ahead of being required to.