What belongs on an AI checklist to prepare for the AMF guideline?

How to read this checklist

This is a generic, best-practice checklist for an institution getting ready for the AMF AI guideline, which comes into force May 1, 2027. Each item is a named control pattern with one sentence of grounding, drawn from the guideline’s own expectation language (« devrait »). It is not legal advice, and no checklist certifies compliance. The guideline’s binding text is French.

The items

• Designate an accountable person for your AI systems (SIA). The guideline expects senior management to name a « personne imputable » for all of the institution’s AI systems.
• Stand up a centralized AIS inventory (« répertoire centralisé »). Keep a regular record of every model and AI system, with non-negligible-risk systems listed in one inventory.
• Assign and update a risk rating (« cote de risque ») for each system. Rate each AI system’s risk and refresh the rating periodically.
• Govern each system across its lifecycle (« cycle de vie des SIA »). Attach governance to every stage, from design through retirement.
• Run continuous monitoring (« surveillance continue »). Watch systems in production, not only at approval.
• Keep contemporaneous records of all of it. One well-kept record can also serve OSFI E-23, finalized and effective the same day — its model inventory, risk ratings, and lifecycle map answer the same questions the AMF asks.

Where a checklist stops

A generic checklist informs your readiness; it does not do the work or prove it was done. The versioned, record-integrated governance instruments library — the same items as procedures and templates you can adopt — ships inside AI Auditability, where adopting each instrument lands in the record as dated evidence. Agentica, an AI Governance company in Montréal, keeps that record continuously, so one record can serve both the AMF and OSFI E-23.

Agentica provides AI-governance evidence and risk intelligence. It does not constitute legal advice or compliance certification.