> 📥 The full essay is available as a PDF (Danish, 8 pages). Download it at the top of the page — summer reading long enough for a flight, concrete enough to make the first board meeting after the break look different.
1. AI has already flown into the boardroom
Two-thirds of US board members already use AI in their board work. That's from Diligent Institute's latest Director Confidence Index. And nearly half of them do it in free consumer tools like ChatGPT and Gemini.
Picture it, now that the summer holiday is around the corner: a board member on their way to the airport, dropping a confidential several-hundred-page board pack into a free chatbot for a flight summary. A US governance expert in Corporate Board Member uses that example as a real risk, not a caricature.
The boardroom in numbers:
- 2/3 of US board members already use AI in board work (Diligent Institute & Corporate Board Member, 2025).
- 66% of board members report limited or no AI knowledge and experience (McKinsey global survey, 2025).
- 39% of Fortune 100 companies had disclosed board-level AI oversight by 2024 (Stanford / Larcker et al., 2025).
Shadow AI has reached the boardroom — long before anyone got a policy written. Adoption is ahead of both competence and oversight.
2. Two conversations — most firms only have one
Reading across McKinsey, BCG, Deloitte, EY, PwC, KPMG, Korn Ferry, Egon Zehnder, Implement, INSEAD and Stanford, the material falls into two piles.
The first is about the board's oversight of AI. How does the board supervise the company's AI use? Do we need an AI-savvy director? Which committee owns the risk? That's where the bulk of the material sits. It's a conversation I've dedicated a full guide to: "The Board's AI Discipline".
The second is far smaller and far more interesting: AI as amplification of the board's own work. Board work is cognitive work. Pattern recognition, scenarios, weighing risk under uncertainty. Exactly what language models are best at supporting. Implement calls their concept "the cyboard". PwC runs hands-on AI sessions for boards. INSEAD launched an "AI for Boards" program built on interviews with 50+ board chairs from companies like Nestlé, Novo Nordisk and Shell. As far as I can find publicly, that's about it.
3. The best ideas across the firms
Boil it down and four use cases remain:
1. Meeting prep — the most common today. AI summarizes, finds inconsistencies and surfaces implicit assumptions.
2. Real-time strategy testing — what-ifs tested in the meeting instead of three weeks later. Before: consulting house + six weeks. Now: a well-set-up AI + ten minutes.
3. Synthetic perspectives — AI personas challenging consensus. Implement calls it having 100 experts in the room without the cost or egos. IHC has had Aiden Insight as an observer since 2024.
4. Sparring for the chair — the most overlooked, biggest effect per krone. The chair's prep lifts the whole meeting.
INSEAD researchers conclude that every board will eventually have some form of AI member. Whether or not you buy the full version: the direction is clear.
4. Five use cases you won't find in the reports
The four above are good. But when I sit with real boards and their own material, it's five other use cases that move the needle. None of the firms describe them. They require no new tools and no new budgets. Only material you already own, and questions you haven't asked yet.
1. The decision audit. Five years of minutes, proposals and budgets is a data source you already own. One question: which assumptions kept recurring, and which held? Most boards find two or three systematic biases in their own optimism.
2. The framing test of the board packet. Every proposal is an argument: chosen numbers, omitted numbers, an order building to a conclusion. Have the model read it as devil's advocate.
3. Real-time pre-mortem. Frame the task as a story: it's 18 months later, the decision turned out to be a mistake. Write what went wrong. Ten minutes, mid-meeting, while the decision can still change.
4. The disagreement log. Every time the AI analysis and the board decision diverge, log it with a two-line rationale. After a year: an honest picture of where the table beat the model, and vice versa.
5. The agenda audit. Have the model analyze 12–24 months of agendas: backward-looking oversight vs. forward-looking strategy? Which topics have recurred three meetings without a decision?
> Example prompt · The decision audit: "Here are our decision proposals and minutes from the last five years. Find the assumptions that recur across cases, and show me where reality diverged from the proposal. I want patterns, not individual cases."
Common thread: the board's own material in a closed environment.
5. What the glossy frameworks talk around: The power balance
The biggest question isn't technical. It's political. Boards have historically depended on information curated by management. Information asymmetry has been a baseline condition, and the whole governance model is built around it.
AI shifts that. A board with a well-configured AI setup can suddenly challenge management's numbers and assumptions in operational depth. Stanford's David Larcker warns of competing information sets between board and management. Korn Ferry raises the legal question: the business judgment rule has for decades protected directors who rely on management's information. What happens when AI gives them the ability — perhaps the duty — to see past it?
That question belongs on the table before the tools are rolled in. Handled afterwards, it becomes a conflict instead of an agreement.
6. The European reality US playbooks skip
Almost all the material is written for a US audience. A Danish board operates under GDPR and the EU AI Act. Board packs contain personal data, inside information and trade secrets. A director who uses a free consumer chatbot for meeting prep is potentially sharing confidential material with a platform that may use it for training.
The order is non-negotiable: closed environment first. Same AI environment the company itself has approved, with DPAs and EU data boundaries in place. Free platforms are ruled out for board work. Period.
> Key numbers · The oversight job is growing at the same time: On 2 August 2026, most of the EU AI Act applies. Fines go up to €35M or 7% of global revenue for prohibited practices, and up to €15M or 3% for most other breaches. Add shadow use: IBM's 2025 report says one in five breaches involved shadow AI, at an extra ~$670,000 per incident.
7. The order runs all the way up
In "AI for Top Executives" the decisive question was whether the CEO opens the tool themselves in the morning. Here it's the same, only with the chair in the lead. A chair who has prepared a meeting with AI asks different questions than a chair who has read about it. That experience can't be delegated — and it's contagious.
The four Judgment Loop questions can serve as the board's rules of play: What does the model analyze? What does the board decide? What must the board see to judge? And what happens when model and board disagree? The last one is the most important. Disagreement is the value. An AI that never challenges your assumptions is an expensive echo of what you already believed.
> The principle across all three articles: AI informs, humans decide. In the boardroom, that sentence is the entire allocation of responsibility. Decision responsibility can't be outsourced — not to a consulting house, not to a model. It stays at the table.
8. How we approach it at TryZone
1. Rules first. A one-page AI policy for the boardroom. Closed environment, no free platforms, AI informs and humans decide, plus an explicit agreement with management. The power-balance question in the open.
2. Then meeting prep. Lowest risk, fastest win. If some at the table are brand new, we start one level earlier with the ten leader prompts from "Generative AI Tools for Leaders".
3. Then real-time analysis and strategy testing. Framing test and pre-mortem live here.
4. Finally synthetic perspectives. The most powerful and the most sensitive.
Along the way we build the disagreement log, and once a year we run the decision audit and the agenda audit. We build within GDPR and the EU AI Act from session one instead of retrofitting compliance at the end.
9. Three questions for your first board meeting after summer
1. Who among us already uses AI in preparation, and in which tools? (The answer usually surprises.)
2. Do we have an agreement with management on how the board may use AI on management's material?
3. If we could test one strategic what-if in real time at the next meeting, which would it be?
The third question is the fun one. It's also where the work starts. Enjoy the summer. Take the article on the plane instead of the several hundred pages. And ask question one out loud at the first meeting back.
> ⚠️ Disclaimer. General practical guidance — not legal or compliance advice. Consult your compliance or legal team and your DPO, and verify dates and status against primary sources (EUR-Lex, European Commission, Datatilsynet). Snapshot as of July 2026.
📬 Subscribe to the newsletter "AI, Built Human" on Substack — weekly insights on AI in practice.
Stefano Vincenti · GenAI Strategist & Architect · External Lecturer, IT University of Copenhagen & DIS Copenhagen · Cofounder & CTO BotTellMe · Partner, TryZone · aitrainer.dk