OSFI Guideline E-23: Model Risk Management
Canada, federal · Supervisory guidance · Published, effective May 1, 2027
- Canada
- Financial Services
© 2026 Medoya LLC
Powered by voilamia
AI Regulation Tracker
Privacy requirements and sector expectations for Canadian AI deployments.
Resources › AI Regulation Tracker › Canada
Overview · United States · European Union · Canada · Standards & Frameworks
May 1, 2027
OSFI E-23 · Supervisory effective date
Canadian AI deployments can engage federal and provincial privacy laws alongside sector-specific supervisory expectations. Start with the organization, activity, personal-data flows, and sector involved. This page highlights selected requirements relevant to technical assessment and deployment.
In each entry, “What this means for your AI deployment” and “Evidence you may need” describe assessment work Medoya suggests. They do not mean that every listed test or document is expressly required by the cited law.
This resource provides general information about selected regulations and technical considerations. It is not legal advice or a determination that a particular system complies. Check the linked official sources and obtain advice for your circumstances.
Assess what an agent can access and do, test relevant attack paths, and document findings and limitations.
Examine how a vendor's product fits the intended use, including data flows, deployment options, available evidence, and open questions.
Review trust boundaries, permissions, human approvals, logging, and recovery arrangements before deployment or expansion.
Coverage can differ by province, activity, and type of information. Check federal, provincial, and sector-specific rules together, including the qualifications and exclusions identified in each entry.
Jurisdiction: Canada, federal · Type: Law or regulation (federal law) · Status: Applicable
Topics: Privacy · Last verified: October 29, 2026 · Last updated: October 29, 2026
Who it may affect. Private-sector organizations and activities within PIPEDA's scope, including relevant federally regulated businesses and cross-border commercial activity.
Summary. PIPEDA addresses accountability, appropriate purposes, consent where required, collection and retention limits, safeguards, and access. Substantially similar provincial laws can govern relevant intraprovincial activities instead. Some provinces also have substantially similar health-information laws.
Watch item (proposed). Bill C-36, which would enact the Protecting Privacy and Consumer Data Act and amend PIPEDA, was introduced in the House of Commons on June 15, 2026 and is awaiting second reading. It is not law; PIPEDA continues to apply.
What this means for your AI deployment. Determine coverage by activity and data flow. Review personal-data use, processing arrangements, retention, deletion, and access controls.
Evidence you may need. Data inventories, documented purposes, vendor arrangements, retention settings, and access-test records.
Scope and limitations. One organization can have obligations under more than one privacy law. PIPEDA is not the only law for every Canadian AI deployment.
Official sources (checked October 29, 2026)
Related services: AI Vendor Assessment · AI Architecture Review & Design
Jurisdiction: Canada, Quebec · Type: Law or regulation (provincial law) · Status: Applicable
Topics: Privacy, Automated decisions · Last verified: October 29, 2026 · Last updated: October 29, 2026
Who it may affect. Enterprises and personal information covered by Quebec's Act respecting the protection of personal information in the private sector (chapter P-39.1).
Summary. Section 3.3 requires privacy impact assessments for specified information-system projects. Section 17 requires a privacy impact assessment and a written agreement before communicating personal information outside Quebec, and applies the same rule when an enterprise entrusts collection, use, communication or keeping of the information to someone outside Quebec. Section 12.1 requires notice and related rights for decisions based exclusively on automated processing, including an opportunity to submit observations to a staff member able to review the decision.
What this means for your AI deployment. Map processing locations and vendors, involve the privacy function early, and review notice, information-request, correction, and review workflows.
Evidence you may need. Privacy assessments, data-flow maps, vendor safeguards, agreements, and workflow tests.
Scope and limitations. Section 3 excludes public bodies, information held on a public body's behalf, and health and social services information within the meaning of Quebec's Act respecting health and social services information (chapter R-22.1) held by or for a health and social services body. Confirm the applicable regime before applying this Act to healthcare data.
Open point. Section 17 speaks of communication outside Quebec, not outside Canada. How it applies to transfers to other provinces is not settled in this entry; confirm with your privacy advisers.
Official sources (checked October 29, 2026)
Related services: AI Vendor Assessment · AI Architecture Review & Design
Jurisdiction: Canada, federal · Type: Supervisory guidance (OSFI guideline) · Status: Published, effective May 1, 2027
Topics: Financial Services · Last verified: October 29, 2026 · Last updated: October 29, 2026
Who it may affect. All federally regulated financial institutions, including foreign bank branches and foreign insurance company branches, to the extent consistent with their branch obligations under OSFI Guideline E-4.
Summary. Published on September 11, 2025, revised E-23 sets proportionate model-risk expectations, including for AI/ML and third-party models. It addresses model identification, risk assessment, inventory, governance, independent review, deployment, and monitoring. Its application is risk-based.
Key dates
What this means for your AI deployment. Understand the model and vendor inventory, deployment controls, dependencies, limitations, and monitoring responsibilities.
Evidence you may need. Model and vendor records, dependency maps, review results, change history, and deployment-test records.
Scope and limitations. This is financial-sector supervisory guidance. A scoped security assessment addresses part of the evidence; full model validation also concerns conceptual soundness and performance. Governance expectations are proportionate to risk.
Official sources (checked October 29, 2026)
Related services: AI Vendor Assessment · AI Architecture Review & Design
Medoya reviews AI agent behavior, vendor evidence, and architecture within an agreed technical scope. We document findings, limitations, and practical next steps so your team can decide what to change and what needs further review.
Select an entry to read its scope, dates, evidence examples, and official sources below.
Canada, federal · Supervisory guidance · Published, effective May 1, 2027
Canada, Quebec · Law or regulation · Applicable