A plain-language guide By Nick Jang, founder · updated 1 November 2026
The AMF’s AI and model risk guidelines, explained.
Quebec’s Autorité des marchés financiers has a model risk guideline in effect since June 2025 and an AI guideline that takes effect on 1 May 2027. Here is what they ask, in plain words, and how they sit beside OSFI’s E‑23. Read the AI guideline and the model risk guideline on the AMF’s site.
01 At a glance
- Who issues them
- The Autorité des marchés financiers (AMF), Quebec's financial sector regulator.
- Who they apply to
- The financial institutions the AMF supervises: insurers, financial services cooperatives, trust companies and other deposit institutions.
- Model risk
- The Model Risk Management Guideline, published June 2025 and in effect.
- AI
- The Guideline for the Use of Artificial Intelligence, final text updated 24 April 2026, in effect 1 May 2027.
- How they fit
- AI systems rely on models, so the AI guideline adds to the model risk guideline rather than replacing it.
- Proportionality
- Both are applied by the nature, size and complexity of the institution's activities and its risk profile.
02 The AI guideline’s five areas
The headings are the AMF’s own; the lines beside them are in our words.
| The AMF's heading | What it asks, in short |
|---|---|
| Institution-wide governance | AI governed across the whole institution, not one business line at a time. |
| Institution-wide risk management | The material risks of AI use managed rigorously across the institution, so it has one complete view of them. |
| Risk-based AIS classification | Each AI system classified by its risk, and managed according to that class. |
| The AIS lifecycle | The governance documents each stage of an AI system's lifecycle needs. |
| Fair treatment of clients | Sound commercial practices toward clients when the institution uses AI. |
03 The AMF beside OSFI E‑23
- Dates
- OSFI's E‑23 and the AMF's AI guideline both take effect on 1 May 2027; the AMF's model risk guideline is already in effect.
- Who
- OSFI supervises federally regulated institutions; the AMF supervises Quebec's. A group with both kinds of entity answers to both.
- Shape
- OSFI folds AI into one model risk guideline, E‑23. The AMF keeps a model risk guideline and adds a separate AI guideline on top.
- In common
- An inventory, risk-based treatment, a governed lifecycle, accountability, and proportionality.
04 Where MINERVA fits
MINERVA keeps AI systems and models in one inventory with their owners, a risk-based rating that sets how often each is reviewed, a lifecycle where every move is decided by a named person and never by the system’s own developer, and evidence an examiner can verify offline, in French and English. It ships rule packs for OSFI E‑23 and AI risk, not yet one written from the AMF’s text. It does not make an institution compliant. See OSFI and AMF software and the E‑23 platform.
05 Questions people ask
What is the AMF's Guideline for the Use of Artificial Intelligence?
The Quebec regulator's expectations for sound and prudent management and sound commercial practices when financial institutions use AI. It covers institution-wide governance, institution-wide risk management, risk-based classification of AI systems, the AI system lifecycle, and fair treatment of clients.
When does the AMF's AI guideline take effect?
On 1 May 2027. The AMF updated the final text on 24 April 2026.
Is the AMF's model risk guideline already in effect?
Yes. The AMF's Model Risk Management Guideline, published in June 2025, is in effect. It covers every stage of a model's life, from the reason for building it to decommissioning, as well as governance and the model risk management framework.
What counts as an AI system for the AMF?
The AMF uses the OECD's definition: a machine-based system that infers from its inputs how to produce outputs such as predictions, content, recommendations or decisions that can affect physical or virtual environments, with varying autonomy and adaptiveness.
Does the AI guideline apply only to AI that handles client records?
No. The AMF says its expectations apply to any use of AI systems by the institution, whether or not it relates to client records.
How is it different from OSFI E-23?
OSFI folds AI into one model risk guideline, E-23, in effect 1 May 2027. The AMF has a model risk guideline already in effect and a separate AI guideline from 1 May 2027. Both ask for an inventory, risk-based treatment, a governed lifecycle and clear accountability.
Does MINERVA support the AMF's guidelines?
MINERVA keeps the record they ask about: AI systems and models in one inventory, a risk-based rating, a lifecycle where each move is decided by a named person, owners and evidence, in French and English. It ships rule packs for OSFI E-23 and AI risk, not yet one written from the AMF's text, and it does not make an institution compliant.
Read more: OSFI E‑23 · OSFI’s AI bulletin · Agentic AI governance · AI vendor risk. A plain-language summary, not legal advice; the guidelines on the AMF’s site are the source.