Board Oversight of AI Risk
Board oversight of AI applies directors' duties of care and loyalty to a fast-moving source of opportunity and risk. Directors should set clear oversight structures, ensure accurate AI disclosure free of AI washing, address shadow AI and confidentiality, decide how AI tools may be used for board meetings, and ask management the right questions.

Key takeaways
- CBCA s. 122(1) requires directors to act in good faith and with reasonable care, diligence and skill.
- Oversight can sit with the full board or a designated committee, but must be explicit.
- CSA Staff Notice 11-348 addresses AI disclosure and warns against AI washing.
- Shadow AI can compromise confidential information and privilege.
- AI note-takers in board meetings raise discovery, privilege and accuracy issues.
- Regular, candid management reporting is the core of effective oversight.
Why board oversight of AI is now a governance priority
Artificial intelligence has moved from pilot projects into core operations, products and decision-making. That shift brings it squarely within the board's remit. Board oversight of AI is not about directors becoming technical experts; it is about applying familiar governance duties to a new source of opportunity and risk: strategy, risk management, disclosure, controls, culture and compliance. Shareholders, regulators, insurers and employees increasingly expect directors to be able to explain how the organization uses AI and how it manages the risks.

This insight explains the legal foundations of directors' duties as they apply to AI, oversight structures, AI in disclosure, shadow AI and confidentiality, the use of AI tools to record board meetings, and the questions boards should ask management. It is general information, not legal advice.
Directors' duties and AI
Under section 122(1) of the Canada Business Corporations Act (CBCA), every director and officer must act honestly and in good faith with a view to the best interests of the corporation, and exercise the care, diligence and skill that a reasonably prudent person would exercise in comparable circumstances. Section 122(1.1) allows directors, when considering the corporation's best interests, to consider the interests of shareholders, employees, retirees and pensioners, creditors, consumers and governments, the environment and the long-term interests of the corporation. Provincial corporate statutes contain similar duties.
Applied to AI, the duty of care calls for an informed, documented process: understanding where AI is material to the business, ensuring management has appropriate systems to identify and manage AI risks, and following up when red flags appear. The CBCA's good-faith reliance defences in section 123 protect directors who rely on financial statements and on reports of professionals whose profession lends credibility to their statements, but reliance must be reasonable and informed. A board that never asks about AI, in a business where AI is material, will find it harder to show it exercised reasonable care.
Oversight structures
There is no single right structure for board oversight of AI. Common approaches include:
| Model | When it fits | Considerations |
|---|---|---|
| Full board | AI is central to strategy or the business model | Ensures strategic discussion; needs regular education |
| Audit or risk committee | AI is mainly a risk and controls issue | Integrates with enterprise risk and internal audit; risk of overload |
| Technology or innovation committee | Large technology-intensive organizations | Deeper expertise; needs clear reporting lines to the full board |
| Shared mandate | AI affects several committees | Requires explicit allocation to avoid gaps, for example compensation committee for workforce impacts |
Whatever the model, committee charters should state who oversees AI, how often management reports, and what triggers escalation. Board skills matrices increasingly include technology and AI expertise, supplemented by director education and access to independent advisers where the board considers it useful.
AI in disclosure and MD&A
For reporting issuers, AI raises disclosure questions. In CSA Staff Notice and Consultation 11-348 (5 December 2024), the Canadian Securities Administrators set out how securities laws apply to AI use by market participants, including non-investment fund reporting issuers. The notice discusses disclosure in the MD&A and annual information form of how AI systems are used, whether they are developed internally or supplied by a third party, the material risks AI presents and the governance, risk management, controls and procedures used to mitigate them, the impact on the business, and the material factors behind forward-looking information about AI.
The CSA encourages entity-specific risk disclosure rather than boilerplate, including context on how the board and management assess and manage AI risks. It also warns against "AI washing": inaccurate, false, misleading or embellished claims about AI capabilities designed to attract investors, which may constitute a misrepresentation under securities law. Audit committees should ensure that AI-related statements in disclosure, investor presentations and marketing are accurate, substantiated and consistent.
Ask management to maintain a short register of public statements about the company's AI capabilities, with the evidence supporting each. Review it before major disclosure filings and investor events. It is a simple control against AI washing claims.
The AI risks boards should understand
- Strategic risk: competitors' use of AI, disruption of business models and investment allocation.
- Legal and regulatory risk: privacy, human rights, consumer protection, intellectual property, sector rules and foreign AI laws.
- Operational risk: errors, hallucinations, model drift, dependency on vendors and resilience.
- Security risk: data leakage, prompt injection, model manipulation and AI-enabled attacks.
- Reputational and ethical risk: biased outcomes, misuse and erosion of trust.
- Workforce risk: skills, change management and labour relations.
Shadow AI and confidentiality
Shadow AI refers to employees using AI tools that the organization has not approved, often consumer generative AI services. The risk is that confidential business information, personal information or privileged legal advice is entered into tools whose terms allow the provider to retain or use inputs. For boards, the issue is twofold: whether management has an acceptable use policy and technical controls in place, and whether directors themselves follow them. Directors who paste board materials into public AI tools to summarize them can compromise confidentiality and privilege. See our insight on generative AI policies.
AI note-takers and recording board meetings
AI meeting assistants that record, transcribe and summarize discussions are spreading quickly, including into board and committee meetings. They raise several governance and legal issues:
- verbatim transcripts and recordings may be producible in litigation or regulatory investigations, unlike traditional minutes that record decisions and key considerations;
- recording privileged discussions with counsel through a third-party service can put privilege at risk;
- AI summaries can be inaccurate or omit nuance, and should never replace approved minutes;
- vendors may store recordings outside Canada or use them to improve their services; and
- participants may need to be informed and consent to recording.
Many boards are adopting a policy that prohibits recording or AI transcription of board and committee meetings unless specifically approved, with the corporate secretary remaining responsible for minutes.
Setting an AI risk appetite
A risk appetite statement for AI helps translate board oversight of AI into practical guidance for management. It can identify uses the organization will not pursue, uses that require board or committee approval, and the tolerances that apply to error rates, customer harm, regulatory exposure and vendor concentration. It should be consistent with the enterprise risk appetite and reviewed as technology and regulation evolve. Management can then design approval thresholds and controls that reflect the board's stated tolerance, and report exceptions. Without an articulated appetite, decisions about AI tend to be made project by project, and the board learns of significant risks only after they have materialized.
Director education and expertise
Directors do not need to understand how models are trained, but they do need enough fluency to ask informed questions and judge the answers. Effective boards schedule periodic education sessions with management and outside experts, include AI in board evaluations and skills matrices, and consider whether specialist expertise is needed on the board or through an advisory panel. Education should cover the company's actual AI use, not just general trends. Directors should also be wary of relying on a single "AI director": board oversight of AI is a collective responsibility, and every director must be able to engage with the issues that affect the company's strategy and risk.
Using AI tools for board work
Some boards are exploring approved AI tools to summarize board packs, compare documents or prepare questions. Used carefully, within a secure enterprise environment with clear retention and confidentiality settings, these tools can help directors prepare. The board should approve which tools may be used, for which materials, and on what terms, and directors should remember that they remain responsible for reading and understanding the materials themselves. Reliance on an AI summary is not reliance on management or professional advice for the purpose of statutory defences.
What good management reporting looks like
Boards should receive regular, concise reporting that includes the AI inventory and material use cases, high-risk systems and their assessments, incidents and near misses, regulatory and litigation developments, key vendor dependencies, progress on the AI governance program, and metrics on adoption and value. Reporting should be candid about limitations and uncertainties, and should distinguish clearly between AI projects that are in production and those still being explored or tested. See our insight on AI governance programs.
Questions boards should ask management
- Where is AI material to our strategy, operations and products today, and in three years?
- Do we have a complete inventory of AI systems, including vendor and employee tools?
- How do we classify AI risk, and which systems are high risk?
- Who is accountable for AI governance, and how are decisions escalated?
- How do we test for accuracy, bias, security and robustness before and after deployment?
- What do our contracts with AI vendors say about data, IP, liability and change management?
- Are our public statements about AI accurate and supported?
- What incidents or complaints have we had, and what did we learn?
- How are we preparing employees for AI-driven change?
- What is our risk appetite for AI, and where are we outside it?
Using recognized standards
Boards can ask management to benchmark the AI program against recognized frameworks, such as ISED's voluntary code of conduct for advanced generative AI, ISO/IEC 42001 and the NIST AI Risk Management Framework. Benchmarking does not create legal obligations by itself, but it provides a structured basis for oversight and for explaining the program to stakeholders.
Liability exposure
Claims against directors for AI failures in Canada remain largely untested. The most likely sources of exposure are securities misrepresentation claims based on AI-related disclosure, oppression remedy claims, regulatory proceedings in regulated sectors, and claims following privacy or security incidents. Directors' and officers' insurance policies should be reviewed for AI-related exclusions and coverage of regulatory investigations. A documented, informed oversight process is the best protection: minutes should record the information considered, the questions asked and the follow-up required, without becoming a verbatim transcript.
Getting started with board oversight of AI
For boards beginning this work, a practical sequence is to assign oversight responsibility in committee charters, hold an education session on the company's AI use and risks, request an initial inventory and risk assessment from management, agree a reporting cadence and metrics, review disclosure controls for AI statements, and adopt a policy on AI tools for board materials and meetings. Effective board oversight of AI then becomes part of the normal governance calendar rather than an annual special topic.
How TRW & Co helps
We advise boards and management teams on board oversight of AI, including committee mandates, AI governance frameworks, disclosure review for AI washing risk, policies on AI tools for board work, and director education. See our corporate governance and artificial intelligence pages or contact us.
Frequently asked questions
Do directors have specific legal duties regarding AI?
There is no AI-specific duty in Canadian corporate law, but the general duties in CBCA s. 122(1) and provincial equivalents apply. Where AI is material, directors should ensure management has appropriate systems to identify and manage AI risks and should follow up on red flags.
Which committee should oversee AI?
It depends on how central AI is to the business. Options include the full board, the audit or risk committee, a technology committee or a shared mandate, provided responsibilities are explicit in charters.
What is AI washing?
The CSA uses the term for inaccurate, false, misleading or embellished claims about AI use designed to attract investors. Such claims may constitute a misrepresentation under securities law.
Should we use AI note-takers in board meetings?
Many boards prohibit them unless specifically approved, because verbatim records may be producible in litigation, privileged discussions may be exposed, and summaries can be inaccurate.
What should management report to the board on AI?
The AI inventory and material uses, high-risk systems, incidents, regulatory developments, vendor dependencies, governance program progress and adoption metrics.
Sources
General information only, not legal advice. References to statutes, courts and authorities describe published law and practice at the date shown.


