AI Governance Programs: ISO 42001, NIST and the Canadian Code
With no federal AI statute in force, AI governance programs are how Canadian organizations manage AI risk under existing law and stakeholder expectations. A practical program combines an AI inventory, risk tiers and impact assessments with controls drawn from the federal voluntary code, ISO/IEC 42001, the NIST AI RMF and privacy regulators' principles.

Key takeaways
- AIDA died with Bill C-27 in January 2025; existing laws still apply to AI.
- The federal voluntary code on generative AI sets measures under six principles.
- ISO/IEC 42001 provides a certifiable AI management system structure.
- NIST's AI RMF is organized around govern, map, measure and manage.
- Risk tiers focus assessments and approvals on high-impact uses.
- Vendor controls and monitoring are essential parts of any program.
Why Canadian organizations need AI governance programs
Canada does not yet have a federal statute regulating artificial intelligence in the private sector. The Artificial Intelligence and Data Act proposed in Bill C-27 died when Parliament was prorogued in January 2025. That gap does not mean AI is unregulated. Privacy law, human rights law, consumer protection, sector regulation, employment law, product liability and contract all apply to AI systems, and customers, investors and insurers increasingly ask how AI risk is managed. Well-designed AI governance programs give organizations a structured way to answer those questions and to adopt AI with confidence.

This insight sets out the building blocks of an operational program, drawing on the federal voluntary code for generative AI, ISO/IEC 42001, the NIST AI Risk Management Framework and the privacy regulators' principles for generative AI. It is general information, not legal advice.
The reference frameworks
| Framework | What it is | How it helps |
|---|---|---|
| Voluntary Code of Conduct on generative AI (ISED, September 2023) | Commitments by signatories developing or managing advanced generative AI systems | Canadian baseline measures on accountability, safety, fairness, transparency, oversight and robustness |
| ISO/IEC 42001:2023 | International standard for an AI management system | Certifiable management system structure: policy, roles, risk assessment, controls, audit and improvement |
| NIST AI Risk Management Framework 1.0 (January 2023) | Voluntary US framework organized around govern, map, measure and manage functions | Detailed practices and a generative AI profile (NIST-AI-600-1, July 2024) |
| OPC and provincial principles for generative AI (December 2023) | Privacy regulators' principles for responsible generative AI | How privacy law applies to AI development and use |
Most organizations do not need to choose one framework. A practical approach uses ISO/IEC 42001 or NIST as the structural backbone, maps Canadian expectations from the voluntary code and privacy principles onto it, and adds sector-specific requirements.
The federal voluntary code
ISED's Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems, launched in September 2023, sets out measures under six principles: accountability, safety, fairness and equity, transparency, human oversight and monitoring, and validity and robustness. It distinguishes developers from managers of systems and applies additional measures to systems available for public use. Examples include implementing a comprehensive risk management framework with policies, procedures and training; assessing reasonably foreseeable adverse impacts; curating training data to manage bias; clearly identifying systems that could be mistaken for humans as AI; monitoring for harmful uses after deployment; red-teaming; and assessing cyber security risks such as data poisoning.
The code is voluntary and binds only signatories, which include Canadian AI developers, banks, telecom and technology companies. But it is a useful statement of what Canadian government considers responsible practice, and it is often referenced in procurement questionnaires.
ISO/IEC 42001 and certification
ISO/IEC 42001:2023 specifies requirements for establishing, implementing, maintaining and continually improving an AI management system. It follows the familiar structure of other ISO management system standards, such as ISO/IEC 27001 for information security, which makes it easier to integrate with existing compliance programs. Key elements include an AI policy, defined roles and responsibilities, AI risk assessment and treatment, AI system impact assessment, controls over the AI system life cycle and data, supplier management, internal audit and management review. Organizations can seek third-party certification, which some customers are beginning to request from AI vendors.
The NIST AI Risk Management Framework
The NIST AI Risk Management Framework, released on 26 January 2023, is intended for voluntary use and is organized around four functions: govern, map, measure and manage. Its companion playbook offers suggested actions, and NIST released a Generative AI Profile (NIST-AI-600-1) in July 2024. Canadian organizations with US operations or customers often use NIST because it is widely recognized and detailed, and because it maps well to ISO/IEC 42001.
Building the AI inventory
Every AI governance program starts with knowing what AI is in use. The inventory should capture internally developed models, AI features embedded in vendor software, generative AI tools used by employees, and AI used by service providers on the organization's behalf. For each system, record the business owner, purpose, model and vendor, data used for training and operation (including personal information), the decisions or outputs it influences, the users affected and the controls in place. Shadow AI, meaning tools adopted without approval, is a common gap; procurement controls, network monitoring and employee surveys help surface it.
Risk tiers
Not every AI use needs the same scrutiny. A tiered model focuses effort where it matters:
- Prohibited or restricted uses, such as uses the organization decides are inconsistent with its values or legal obligations;
- High risk, for example AI that makes or materially influences decisions about individuals' employment, credit, insurance, housing, health care or access to services, or that operates in safety-critical contexts;
- Medium risk, such as customer-facing chatbots and content generation with human review;
- Low risk, such as internal productivity tools with no personal or confidential data.
Criteria should consider impact on individuals, scale, autonomy, sensitivity of data, explainability and regulatory exposure. The tier then determines the required assessments, approvals and monitoring.
AI impact assessments
For higher-risk systems, an AI impact assessment documents the purpose, data, model, testing, risks and mitigations before deployment. It should integrate the privacy impact assessment, since privacy regulators expect organizations to assess generative AI for privacy risks, and Quebec requires privacy impact assessments for information system projects involving personal information. The principles for generative AI issued by the OPC and its provincial and territorial counterparts address legal authority and consent, appropriate purposes, necessity and proportionality, openness, accountability, individual access, limiting collection, use and disclosure, accuracy and safeguards. Public sector bodies will also recognize the algorithmic impact assessment used under the federal Treasury Board directive on automated decision-making.
Do not build a separate AI assessment that ignores privacy, security and human rights reviews already in place. A single intake form with branching questions, routed to the right specialists, saves time and reduces the risk that a high-risk use slips through.
Vendor controls
Much of an organization's AI risk sits with vendors. AI governance programs should extend procurement due diligence to AI-specific questions: training data sources, use of customer data for training, model change management, testing for bias and accuracy, security controls, incident notice and audit rights. Contracts should reflect the answers, and higher-risk vendors should be reassessed periodically rather than only at onboarding. See our insight on AI vendor contracts.
Monitoring and incident management
AI systems change over time as data, models and usage evolve. Post-deployment monitoring should track performance, drift, complaints, errors and harmful outputs, with thresholds that trigger review or suspension. An AI incident process, aligned with security and privacy incident response, should define how issues are reported, triaged, fixed and documented. The voluntary code's expectation that developers maintain a database of reported incidents is a useful benchmark.
Roles, policies and training
Effective AI governance programs assign clear accountability: an executive sponsor, a cross-functional AI committee (legal, privacy, security, risk, data science, business), system owners and an independent review function. Core policies typically include an AI policy, an acceptable use policy for generative AI tools, model risk standards and procurement requirements. Training should be role-based, from general awareness for all staff to deeper training for developers and approvers. See our insight on generative AI policies for employees.
Fairness, human rights and explainability
Canadian human rights statutes prohibit discrimination in employment, services, housing and other areas, regardless of whether a human or an algorithm made the decision. AI used to screen job applicants, set prices, assess credit or allocate services can reproduce bias in training data or proxies for protected grounds. AI governance programs should require bias testing proportionate to risk, using appropriate metrics and representative data, before deployment and at intervals afterward.
Where AI informs decisions about individuals, organizations should be able to explain the main factors in plain language, provide a route to human review, and keep records that show how the decision was reached. Quebec's private-sector law already requires organizations to inform individuals when a decision is based exclusively on automated processing and to allow them to submit observations.
Documentation and records
Documentation is what turns good intentions into demonstrable governance. For each higher-risk system, keep the approved impact assessment, testing results, model and data documentation, approvals, vendor diligence, monitoring reports and incident records. Regulators, auditors, customers and courts increasingly ask for this evidence, and it is much harder to reconstruct after a complaint or claim. Well-run AI governance programs treat documentation as part of the system life cycle rather than a separate compliance exercise.
Measuring maturity
Programs evolve. A simple maturity model, from initial (ad hoc decisions) to defined (policies and inventory in place) to managed (risk tiers, assessments and monitoring operating consistently) to optimized (metrics-driven improvement and external assurance), helps leadership set realistic goals. Useful metrics include the percentage of AI systems inventoried and risk-tiered, assessments completed before launch, open high-risk findings, incidents and time to resolution, and training completion. Periodic internal audit or an external readiness review against ISO/IEC 42001 can validate progress and identify gaps before customers or regulators do.
Board reporting
Boards need concise reporting on the AI inventory, high-risk systems, incidents, regulatory developments and the maturity of the program. Directors should also understand which strategic initiatives depend on AI, what risk appetite management has adopted, and how exceptions to the program are approved and tracked. See our insight on board oversight of AI.
Sector-specific expectations
Regulated sectors face additional expectations. Federally regulated financial institutions apply OSFI's model risk and technology risk guidance. Securities regulators have published guidance on the use of AI in capital markets. Health care organizations must consider health privacy statutes and medical device rules for AI tools. Employers must account for human rights law and, in Ontario, disclosure requirements for AI used in hiring under employment standards legislation. Programs should map these overlays onto the core framework.
Getting started with AI governance programs
- Appoint an executive sponsor and form a cross-functional committee.
- Build an initial AI inventory, starting with customer-facing and high-impact uses.
- Adopt a risk-tiering method and an integrated intake and assessment process.
- Choose a backbone framework, such as ISO/IEC 42001 or NIST, and map Canadian expectations onto it.
- Update procurement and vendor contracts for AI.
- Publish an acceptable use policy and train staff.
- Set up monitoring, incident management and board reporting.
How TRW & Co helps
We help organizations design and operate AI governance programs suited to Canadian law and their risk profile. Our work includes AI inventories and risk tiering, impact assessment templates, policies, vendor due diligence and contract terms, ISO/IEC 42001 readiness and board briefings. See our artificial intelligence and privacy and data protection pages or contact us.
Frequently asked questions
Is there a federal AI law in Canada?
Not for the private sector. The Artificial Intelligence and Data Act in Bill C-27 died on prorogation in January 2025. Privacy, human rights, consumer protection and sector rules continue to apply to AI.
Should we pursue ISO/IEC 42001 certification?
It depends on your customers and risk profile. Certification can help AI vendors demonstrate maturity, while other organizations may use the standard as a framework without certifying.
Does the voluntary code apply to us?
It binds only signatories, but it describes measures the federal government considers responsible practice for advanced generative AI and is often used as a benchmark.
What belongs in an AI inventory?
Internally built models, AI features in vendor products, employee generative AI tools and AI used by service providers, with owners, purposes, data, affected individuals and controls.
How do AI and privacy assessments fit together?
Use an integrated intake and assessment so privacy impact assessments, security reviews and AI-specific risk analysis are done together, consistent with privacy regulators' generative AI principles.
Sources
General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.


