AI and Boards: It Is Time to Get Practical
Written by Mauro da Cunha, Skytop Contributor / July 22, 2026
Mauro R. da Cunha has served as an independent director for more than 25 years, He currently is a director on the boards of Hypera Pharma, Klabin, and Tupy, serving as Chairman of the Audit Committee of the latter. He is also the founder of Engage.MC, an advisory firm focused on corporate governance, strategy and stewardship.
Directors are under growing pressure to understand artificial intelligence. That urgency has produced an expanding market of conferences, advisers and technology providers promising to explain what AI means for boards. Much of the discussion, however, remains either abstract or promotional. How should directors use AI in the performance of their own duties? The question deserves a more substantive explanation.
Moving Boards into the Intelligence Economy
Ignoring these tools is not a credible long-term position. AI can improve access to information, reduce administrative burdens and help directors navigate increasingly complex board materials. Yet uncritical adoption is equally problematic. Even where a product has been reviewed by IT, cybersecurity and legal teams, blind adoption might make boards worse, not better.
Most AI tools currently marketed to boards fall into three categories. The first is the AI notetaker, which records meetings, produces transcripts or summaries. The second is the generative-AI assistant, typically a chat interface that searches board materials and answers questions such as: “Summarize the papers for the next meeting” or “Compile the board’s recent discussions on executive remuneration.” The third, and most ambitious, is the “AI director”: a virtual participant designed to contribute observations, challenge assumptions or recommend courses of action.
The case for adopting all three may appear compelling. Board decisions are consequential, and the cost of software is immaterial relative to the value of better decisions. But cost is not the principal issue. The real questions concern board culture, engagement and accountability.
The Potential for Both Positive and Negative Impact
Consider the AI notetaker. A complete record of boardroom discussion may inhibit candor in much the same way as audio recording. Directors may become less willing to test ideas, voice uncertainty or challenge management if every intervention is preserved and potentially discoverable in litigation or regulatory proceedings. The board may gain administrative efficiency while losing the spontaneity and constructive tension on which effective oversight depends.
A generative-AI assistant presents a different risk. Search and retrieval tools can be extremely useful, particularly when directors need to revisit the history of a complex issue. But using AI to summarize papers for an upcoming meeting may encourage superficial preparation. Directors may arrive having read only a machine-generated digest of management’s own summary. Nuance, dissenting evidence and weak signals can disappear through successive layers of compression.
New Risks to Consider
This risk is not merely individual. If every director relies on the same synthesis, the board may begin from a standardized framing of the issue. That can reinforce groupthink, narrow the range of questions asked and weaken the board’s ability to challenge management or identify alternative courses of action. AI should increase directors’ capacity to engage; it should not become a substitute for engagement.
The virtual AI director raises more fundamental concerns. It may have value as a controlled source of challenge, scenario testing or pattern recognition. But it should not be treated as a decision-maker, nor should directors defer to it as an apparently neutral or all-knowing participant. Its output will reflect its training data, system design, access permissions and prompts.
Holding on to Human Intelligence and Judgement
Accountability remains with human directors, who must exercise independent judgment and cannot outsource their fiduciary responsibilities. Boards should therefore establish explicit protocols before deploying these tools. This takes time and effort, but it is crucial for effective results.
For AI notetakers, access should be limited to the company secretary or governance officer and used only to support the preparation of minutes. Not even individual directors should have access to these documents. Raw recordings, transcripts and AI-generated summaries should be destroyed after the minutes are approved and signed, except in case of legal, regulatory or judicial mandate to the contrary.
For generative-AI assistants, boards should distinguish between retrospective search and prospective preparation. Empowering directors to search materials from previous meetings with efficiency can strengthen institutional memory. Allowing AI to replace the reading of papers for a forthcoming meeting is far more problematic. AI assistants should therefore be barred from accessing materials for forthcoming meetings. They should only be employed as an “enhanced search function” to provide context and history.
Virtual directors should be tested, if at all, in a controlled environment. Early uses might include pre-meeting scenario analysis, identification of missing information or structured challenge exercises. The tool should not vote, be counted for quorum, participate as though it were a director or be presented in a way that obscures who is legally responsible for the decision.
The same discipline should apply when management uses AI to prepare board materials. AI may help anticipate questions, improve structure and identify missing analysis. It may also amplify management’s preferred narrative, suppress alternatives or create an appearance of completeness unsupported by the underlying evidence. It other words, it can increase biases.
Board papers should in fact present realistic options, principal risks, assumptions, counterarguments and the consequences of not proceeding.
Risks to Board Governance
Governance officers have a central role in this framework. They can coordinate with legal, cybersecurity, records-management and business teams; define permissible use cases; monitor vendors; and ensure that AI supports, rather than weakens, board effectiveness. If vendors are not willing to customize their offerings to these protocols, board should consider a different approach, using approved AI engines in a restricted folder in the corporate cloud, instead of the gated board portal.
The objective is not to slow adoption. It is to adopt deliberately and to think independently in term of what can improve the board’s decisions. The right test is not whether an AI tool is innovative, efficient or widely available. It is whether its use improves the board’s access to information while preserving candor, independent judgment, accountability and meaningful engagement.
Boards that apply those principles will be better positioned to benefit from AI without allowing convenience to erode governance.