Skip to main content
Agentica

Answer — AI Governance

What is the difference between OSFI E-23 and the AMF AI guideline?

Two texts, one date: May 1, 2027. OSFI’s E-23 governs model risk management for federally regulated institutions and explicitly includes AI in “model.” The AMF guideline governs the use of AI by institutions authorized in Québec, with an AIS inventory expected. Two vocabularies, one question of proof.

The two texts, side by side

E-23 (OSFI)AI guideline (AMF)
SupervisorOSFI — federalAMF — Québec
Applies toFederally regulated financial institutionsInsurers, financial services cooperatives, trust companies, and deposit institutions authorized in Québec
SubjectModel risk management — “model” explicitly includes AI and machine learningThe use of artificial intelligence
Expected inventoryA comprehensive, accurate, evergreen “model inventory” with risk ratingsA centralized “AIS inventory” for AI systems with non-negligible risk
NatureSupervisory expectations (“should”)Supervisory expectations (“should”); the authoritative text is French
Final sinceSeptember 11, 2025April 7, 2026
In forceMay 1, 2027May 1, 2027

Who answers to both?

Financial groups active in Québec and under federal oversight — a federal insurer with Québec operations, a deposit institution under dual supervision. For them, May 1, 2027 is a double deadline on the same day: two supervisors, two inventory vocabularies, and the same underlying question asked twice — what AI systems do you run, and can you demonstrate it?

Do you need two records?

No — one base of facts is enough, if its collection is continuous and its history tamper-proof. What differs between the two texts is the vocabulary of the rendering: one asks for the inventory in model-risk-management terms, the other in AIS terms. A single well-kept register expresses itself in both.

That is Agentica’s approach: continuously map every AI agent in your business environment, record it in a tamper-proof history, and extract the inventory in the vocabulary of whichever supervisor asks — collect once, prove many times.

Sources