When Directors Use AI: Governing the Board's Own Reading

Board directors using AI to prepare for meetings has quietly become normal, and almost no one has decided whether it should be. Most AI governance work points the board at management’s systems; very little of it asks how the board’s own reading of a board pack is now being shaped by a model before anyone walks into the room.

The gap is not hypothetical. What follows looks at how fast director use of generative AI has grown, why a summary can be accurate and still remove the one fact that matters, where responsibility for the board’s judgment sits once a tool has done the first reading, and what chairs and governance committees can decide now, before the first decision goes wrong for a reason no one will see in the minutes.

How many board directors are using AI in board work?

According to Diligent Institute and Corporate Board Member polling reported in October 2026, 82% of US public-company directors said they had used generative AI in their board work in the past six months, up from 66% nine months earlier. Only 2% had ruled it out entirely.

Governance has not kept pace. The same Diligent article cites its 2026 What Directors Think report, in which 22% of boards said they had processes in place to guide directors’ use of AI. Its Director Confidence Index research found that 54% of directors had no guidance at all, and only 6% reported a formal AI policy specific to the board. Among directors who use AI for board activities, 46% rely on consumer tools.

The pattern is familiar from every earlier technology that arrived in the boardroom: adoption first, policy later. The difference here is that the tool is not only handling logistics. It is doing the reading.

Why is an AI summary of a board pack a governance issue?

A board pack is not just information. It is the record of what management chose to put in front of the board, and the board’s duty is to read it critically: to find the assumption buried in an appendix, the risk qualified in a footnote, the number that changed since the last meeting. That reading is the board’s judgment in its earliest form.

An AI summary changes where that judgment begins. The model decides what is central and what is detail. It compresses the record around the most legible facts, and in a long pack those are usually the facts management put in the executive summary. The footnote that qualifies the base case, the covenant that tightens under one scenario, the dissent recorded in a committee note: these are exactly what a compression is built to drop.

None of this requires the summary to be wrong. As one character in a recent book concludes about an AI-organized file, “The summary was true, but not true enough to decide from.” For a board, that distinction is the whole job.

What happens when a director relies on an AI summary?

Consider a composite case. A board is asked to approve a significant capital commitment for a new production facility. The pack runs to well over a hundred pages. One director, short on time after a week of travel, asks an AI assistant to summarize it and list the key risks. The summary is clean and accurate. It covers the investment case, the timeline, the financing, and the main operational risks.

What it does not surface is a footnote in the financing appendix: the base case assumes a supply agreement that has not yet been signed, and without it the payback period roughly doubles. A second director, who read the pack in full, raises it. The discussion changes. The board asks management to return with the agreement in hand.

The first director did nothing unusual. They used a tool most of their peers now use, in a way no policy prohibited. But for the purposes of that decision, their judgment had been formed by a system no one on the board had chosen, tested or agreed to rely on. Had the second director also used a summary, the board would have approved a different risk than the one it thought it was approving, and the minutes would show nothing out of order.

Where does reliance begin on the Second Map of Work?

An org chart shows where formal authority sits. The Second Map of Work follows four points it cannot show: Formation, where an answer is first shaped; Entry, where a person first meets the output; Reliance, where someone begins acting as if the output is trustworthy; and Consequence, where the work lands on a person. Across the path, it tests two responsibilities: Authority and Ownership.

Applied to the board’s own work, the map is uncomfortable. Formation moves from management’s drafting to the model’s compression. Entry is the moment a director opens the summary instead of the pack. Reliance begins when the director walks into the meeting carrying the summary’s view of the risks as their own. Consequence is the vote, and through it every employee, investor and community the decision touches.

Authority, on this map, is not in question. Directors hold it fully. Ownership is the issue. A director who votes on a summary’s reading still owns the vote, but they no longer authored the reading that produced it. That is a quiet shift in the board’s human authorship, and it happens one director at a time, out of sight of the chair.

How should boards govern directors’ own use of AI?

The answer is not to ban the tools. Used well, they help directors prepare, test management’s assumptions and ask better questions. The answer is to make the board’s reliance a decision rather than a habit.

Decide what AI may and may not do in board preparation. A summary used to navigate the pack is different from a summary used in place of it. A chair can set the expectation that material decisions are read in full, and that AI is a second reader, not a first.

Settle confidentiality and tooling. Board materials are among the most sensitive documents a company holds. Directors should know which tools are approved, where pack content may and may not be uploaded, and who has assessed the risk. Consumer tools used on personal devices are the default today because nothing else has been offered.

Ask management to flag what a summary would miss. Management can mark the assumptions, conditions and dissents that most change the decision, so they cannot be compressed away by any reader, human or machine.

Make reliance visible in the room. For major decisions, a chair can ask directors which material they read in full and which they reviewed through a summary. The question is not an accusation. It tells the board where its collective reading is thin.

Put it on the governance committee’s agenda. Board-level AI use belongs in the board’s own evaluation, alongside attendance, independence and preparation. If the board expects management to govern AI, it should be prepared to show how it governs its own.

Frequently asked questions

Is it acceptable for board directors to use AI to read board packs? Yes, if the board has decided how. AI can help directors navigate long materials and sharpen questions. The risk comes when a summary replaces reading on material decisions without anyone choosing that. Boards should set expectations for when full reading is required and which tools are approved for confidential materials.

What should a board AI policy for directors include? At minimum: approved tools and where board materials may be uploaded, confidentiality rules, which decisions require reading the full pack, how directors disclose reliance on AI summaries for major votes, and who reviews the policy. It should be owned by the governance or nominating committee and revisited as the tools change.

Can AI summaries miss important risks in board materials? Yes, even when they are accurate. Summaries compress around the most prominent facts, which tend to be the ones management already emphasized. Footnotes, conditional assumptions, covenant terms and recorded dissent are precisely the details a compression tends to drop, and they are often what changes a board’s view of risk.

Who is accountable if a board decision relies on an AI summary? The directors are. Using a tool does not transfer fiduciary responsibility to the tool or its vendor. That is why reliance needs to be deliberate: a director who votes on an AI-shaped reading owns the outcome as fully as one who read every page, without having authored the reading behind the vote.

---

Board chairs and governance committees working through how their own boards should use AI can start a conversation. Anker Bioss advises boards on governing judgment in AI-shaped work, and Alder Koten conducts executive searches for leaders who can hold that line. Related reading: Legitimacy Without Accountability: AI in the Board Pack.

Jose J. Ruiz is CEO and Managing Partner of Alder Koten and Chairman of Anker Bioss. He develops these ideas at book length in AI in the Org Chart: A Leadership Guide to Implementing AI Without Losing Human Judgment, Accountability, and Trust, published by Elavant Press and available on Amazon.