What is OSFI Guideline E-23?

Who does it apply to, and from when?

Guideline E-23 applies to all federally regulated financial institutions (FRFIs) — banks, federal insurers, trust and loan companies, including foreign branches with proportionate application. Published in final form on September 11, 2025, it takes effect May 1, 2027 — the same day as the AMF’s AI guideline, which creates a double deadline for groups active in Québec.

It is prudential guidance: it states expectations (“should”), not statutory articles. It is nonetheless the text the supervisor will use.

What does it cover, in its own words?

Three elements structure the text:

• A broad definition of “model.” Any application of theoretical, empirical, or judgmental assumptions or statistical techniques — explicitly including AI/ML methods — that processes input data to generate results. Your AI systems qualify.
• A model inventory. Comprehensive, accurate, and evergreen, under robust controls, covering models whose inherent risk is non-negligible — each carrying its model risk rating.
• A governed lifecycle. From design to retirement, with defined roles (owner, developer, reviewer, approver, user) and monitoring over time.

Where do you start, concretely?

With the question the text implies: do you know which models and AI systems are actually running in your institution, today? The expected inventory is not written the night before the examination — it is kept, continuously, with the evidence to back it. A record started in 2026 reaches May 1, 2027 with a history; a record started in 2027 arrives with a promise.

Agentica maintains that base: continuous mapping of every AI agent in your business environment, recorded in a tamper-proof history, from which the inventory in E-23’s vocabulary is extracted — and, for institutions under Québec jurisdiction, the AIS inventory in the AMF’s vocabulary. Collect once, prove many.