Expertise Won’t Cross the Seam on Its Own

The board side of AI oversight: what directors do with what reaches them

Board oversight of AI depends on the expertise of management, individual directors, and outside advisers. But that expertise does not automatically reach a board’s decisions. It must cross a seam: the point where information and analysis meet director judgment.

This is the first of three articles on that seam. Here, I focus on what the board does with what reaches it. The next article will look at how management translates expertise into decision-ready information. A third will consider who sits at that seam, and whether the board’s composition and working practices make effective challenge possible.

A survey worth reading

A recent Cybersecurity Dive article on BDO’s 2026 Board Survey caught my attention. BDO surveyed more than 200 directors of U.S. public-company boards in June 2026. Eighty percent said their board had a clear framework for overseeing AI adoption and cybersecurity risk. When asked whether the board had the right skills and experience to oversee AI, cybersecurity, and data governance, 36% were very confident, 49% somewhat confident, and 15% not confident. Eighty-eight percent were confident that management had the expertise to identify and integrate emerging technologies and mitigate evolving cybersecurity risk.

It would be tempting to set the 80% framework figure against the 36% who were very confident and call it an overconfidence gap, but the figures do not support that conclusion. The questions were posed differently: one is a yes-or-no measure, while the other reports the top response on a three-point confidence scale. More accurately, 85% expressed at least some confidence in their board’s skills. That is qualified confidence, not evidence or a measured overconfidence gap.

That confidence exists alongside pressure: 57% cited the pace of AI and cybersecurity capabilities as a significant external concern for the next 6 to 12 months. Roughly one-quarter said the growing complexity of board agendas had led them to deprioritize AI and cybersecurity oversight. These findings do not establish that a particular board is overconfident. They raise a more practical question: does the board have the time and information to exercise the oversight it believes it can provide?

Time on the agenda is not the same as time to do the work

At first glance, these results may appear to conflict. NACD’s 2025 Public Company Board Practices and Oversight Survey found that more than 62% of director respondents set aside agenda time for full-board AI discussions. BDO’s 2026 survey found that roughly 25% had deprioritized AI and cybersecurity oversight because of crowded agendas.

Those results are not contradictory. NACD asked whether boards set aside time for AI; BDO asked whether agenda complexity had caused oversight to be deprioritized. The surveys covered different respondents at different times, and “setting aside time” does not address whether a discussion received the planned time or produced the intended work.

For an individual board, the questions are straightforward: When AI is scheduled for discussion, does it get the time and follow-through it needs, or does it get shortened or deferred as other issues arise? What happens to the work that gets displaced?

Where structures give way

I trained as a civil engineer. When a structure is under stress, attention may shift from its beams and columns to the places where they meet. Two strong members on their own do not make a strong connection. The seam between them must be designed for the load it carries, and it can easily be overlooked precisely because what is on either side looks sturdy.

Board oversight has a seam too: the place where management’s expertise meets the board’s judgment. Expertise on one side does not automatically cross, as it must reach decision-makers in a form they can understand, question, and use.

What reaches the board?

PwC’s 2026 Annual Corporate Directors Survey offers another view of this question. Seventy-one percent of directors said their boards need to strengthen their AI skills to provide more effective oversight, and 82% rated the information they receive linking AI outcomes, risks, and business performance as fair, poor, or not provided.

Again, these findings are from a separate survey and are not a matched comparison with BDO’s respondents. They do not show that directors who feel confident about their board’s skills are the same directors who find the information inadequate, nor do they show that boards accept what management gives them. They do point to a tension; a board may have a governance framework and some confidence in its oversight while still needing stronger AI fluency and more decision-useful information.

Read together, these surveys raise a question rather than establish a single pattern within any one board: do directors have enough protected time, AI fluency, and decision-useful information to challenge recommendations? PwC also found that 81% believe their boards could challenge management on corporate strategy more effectively, which is a broader finding, not specific to AI. For an individual board, the question is what happens when the information is incomplete or uncertain: what enables directors to probe the recommendation, test its assumptions, and understand what would cause management to revisit it?

Questions a board can ask itself

· When AI is scheduled for board discussion, how often does it receive the time and follow-through planned?

· What evidence, assumptions, and uncertainties accompany an AI-related recommendation, and what additional information does the board need before it can act?

· What new information or result would cause management to revisit that recommendation?

· Which directors can challenge the analysis, and how does the full board engage when expertise is concentrated in one or two people?

· How does the board test its confidence in its own oversight against its actual practices and decisions?

At SagaciousThink, this is the kind of divergence BoardPulse is designed to surface: where a board’s perceptions of its oversight differ from the practices and decision patterns behind those perceptions. A structured diagnostic or a candid board conversation can help make that divergence visible. Confidence becomes more useful when a board can connect it to evidence and practice.

Next: the management side of the seam

Most of what reaches the board about AI comes from management. The next article will examine how management translates its expertise into information directors can use—including the assumptions, uncertainties, and evidence behind a recommendation.

A later article will consider who sits at the seam: how board composition and interaction affect whether different perspectives are present, heard, and used.

Research note: The BDO, NACD, and PwC findings cited here come primarily from surveys of public-company directors. They provide context for board discussion; they do not establish what is happening in any one boardroom. The surveys also ask different questions, so their percentages should not be read as directly comparable rates.