Federal (Canada) · Office of the Superintendent of Financial Institutions (OSFI)
OSFI Guideline E-23 (model risk)
OSFI Guideline E-23 – Model Risk Management (2027)
OSFI Guideline E-23 is the federal model-risk rule for banks and other federally regulated lenders. The final version was published September 11, 2025 and takes effect May 1, 2027. It expressly covers AI and machine-learning models, so the algorithms behind credit, pricing and fraud decisions need documented oversight and monitoring.
Figures as of November 1, 2026.Mitey, data editor · Evidence checked Status checked against the official sources listed below.
Final, published September 11, 2025. Takes effect May 1, 2027, and expressly covers AI and machine-learning models.
Next milestone
E-23 effective for federally regulated financial institutions
The editor’s read
Why does it matter for AI in banking?
E-23 is the closest thing Canada has to a binding AI rule for banks. It does not ban any use of AI; it makes each federally regulated institution keep an inventory of its models, rate their risk, validate them independently and monitor them once they are live, with AI and machine-learning models explicitly in scope.
Most large institutions in the Index already say their model-risk frameworks cover AI. What will change on May 1, 2027 is that supervisors can test it. For customers, the practical effect is indirect but real: a bank must be able to explain how a model that decides your credit limit or flags your payment was built and checked.

What are the key dates?
Final guideline published
Guideline takes effect
What does it mean for bank customers?
The models your bank uses to approve credit, set prices or flag fraud, including AI and machine-learning models, will need documented oversight, bias checks and ongoing monitoring. You will not see the guideline directly, but it is the main federal rule that makes a bank answer for how its algorithms behave.
What does it mean for banks?
Federally regulated institutions need an enterprise model inventory, risk ratings and lifecycle controls (design, independent review, approval, deployment, monitoring, decommissioning) that explicitly cover AI/ML, explainability and model drift. Requirements scale with the institution's size and complexity, and heavy use of advanced AI is expected to be matched by more mature governance.
Which institutions in the Index does it touch?
Federally regulated institutions: does their model-risk framework already cover AI?.
| Institution | Evidence | Index score |
|---|---|---|
| TD | Framework covers AI/ML · E-23 not named | 84 |
| RBC | Framework covers AI/ML · names E-23 | 81 |
| CIBC | Framework covers AI/ML · names E-23 | 80 |
| National Bank | Framework covers AI/ML · names E-23 | 69 |
| BMO | Framework covers AI/ML · E-23 not named | 57 |
| Scotiabank | AI coverage not stated · names E-23 | 55 |
| EQ Bank | AI coverage not stated · names E-23 | 42 |
| Tangerine | AI coverage not stated · E-23 not named | 21 |
| Coast Capital | Framework covers AI/ML · E-23 not named | 17 |
| Simplii | AI coverage not stated · E-23 not named | 13 |
| Canadian Tire Bank | AI coverage not stated · E-23 not named | 10 |
| Laurentian Bank | AI coverage not stated · names E-23 | 8 |
| Manulife Bank | AI coverage not stated · E-23 not named | 5 |
| Vancity | AI coverage not stated · E-23 not named | 3 |
| Alterna | AI coverage not stated · E-23 not named | 0 |
| Central 1 | AI coverage not stated · E-23 not named | 0 |
| Home Trust | AI coverage not stated · E-23 not named | 0 |
| PC Financial | AI coverage not stated · E-23 not named | 0 |
| Tru (First West) | AI coverage not stated · E-23 not named | 0 |
What does it ask for, and who already publishes it?
Each line paraphrases one expectation from the official sources. Where the Index measures something close, the last column counts the institutions in scope (federally regulated institutions) that publish it, each linked to its source on their profile.
| What it asks | Closest Index criterion | Institutions that publish it |
|---|---|---|
| Treat AI and machine-learning methods as models under the model-risk frameworkThis is the change that brings AI inside the rule. | Says its own model-risk framework covers AI/ML | 6 of 19TD, RBC, CIBC, National Bank, BMO, Coast Capital |
| Keep a model inventory and rate each model’s riskRiskier models, such as credit decisions, get more scrutiny. | Not measured by the Index | — |
| Validate models independently before and after they go liveCatches drift and bias that the builders miss. | Not measured by the Index | — |
| Give the board and senior management clear accountability for model riskSomeone senior answers for the algorithm. | Discloses board or committee oversight of AI | 4 of 19TD, RBC, CIBC, National Bank |
| Name an executive responsible for AINot required by E-23 itself, but the clearest public sign of accountability. | Names an executive accountable for AI | 7 of 19TD, RBC, CIBC, National Bank, BMO, Scotiabank, EQ Bank |
| Disclose readiness for E-23 in annual reportingInvestors and customers can see the bank is preparing. | Names OSFI E-23, or the AMF equivalent for Quebec-regulated institutions | 6 of 19RBC, CIBC, National Bank, Scotiabank, EQ Bank, Laurentian Bank |
The checklist above, counted from the published Index data. Every number can be rechecked from the criteria CSV.
| Institutions in scope | 19Federally regulated institutions |
|---|---|
| Checklist items | 64 matched to an Index criterion |
| Items published, across all institutions in scope | 23 of 76One point for each institution that publishes each measured item |
| Most items published by one institution | RBC: 4 of 4 |
| Share of measured items published | 30%3 of 19 institutions publish every measured item |
What do the key terms mean?
- Model
- Under E-23, any method that turns data into an estimate, forecast or decision, explicitly including AI and machine-learning methods.
- Model risk
- The chance of loss or harm from a model that is wrong, misused or poorly controlled, for example a credit model that drifts after launch.
- Validation
- An independent check, before and after a model goes live, that it works as intended for the data it sees.
What this means for you
- It is final but not yet in effect (from May 1, 2027).
- To see what your own bank has disclosed about AI and your rights today, use What AI touches my account?
- To complain about an AI-assisted decision, start with your bank’s complaint process; our guide on complaining about a bank AI decision has the steps.
Questions about OSFI Guideline E-23 (model risk)
What is the status of OSFI Guideline E-23 (model risk)?
Final, published September 11, 2025. Takes effect May 1, 2027, and expressly covers AI and machine-learning models.
Does E-23 apply to credit unions?
Only to federally regulated ones, such as Coast Capital and Tru Cooperative Bank. Provincially regulated credit unions follow their provincial regulator; Desjardins follows the AMF’s own model-risk and AI guidelines.
Does E-23 give me a right to an explanation?
No. It sets duties for the institution and its supervisor. Rights to an explanation come from privacy law, such as Quebec’s Law 25, and would expand under Bill C-36 if it passes.
Sources (official first)
Change log
Status verified against the official sources: Final, published September 11, 2025. Takes effect May 1, 2027, and expressly covers AI and machine-learning models.
Status as checked on November 1, 2026. Laws and guidance change; check the official source before relying on a date.
