Is it safe for board directors to put board materials into an AI tool?

Only in a tool the company has approved and controls. Board packs carry forecasts, deal names, personnel matters and privileged legal advice, and a personal account on a consumer AI product puts that material under retention terms the director has likely never read. Kaykas-Wolff recommends banning personal accounts for any nonpublic board material.

What should a CEO do about directors summarizing the board pack with AI?

Test the pack before it ships. Kaykas-Wolff suggests someone on management's side run the pack through the kind of tool directors use, compare the summary with what management most needs the board to understand, and rewrite whatever fell out. The usual fix is putting the condition in the same sentence as the number.

Why do AI summaries of board packs mislead directors?

Summarizers keep headline numbers and drop the conditions attached to them, especially when a caveat sits in an appendix pages away from the slide. A director can arrive having absorbed that bookings beat plan without learning that two deals were pulled forward from the next quarter.

Should management provide the board's AI summary or assistant?

Kaykas-Wolff argues it should not. The company should run any board AI tool through its security review, but the board should own the configuration through the chair or corporate secretary, and management should never be able to see the questions directors ask it.

Should AI tools be used on the board's executive session?

No. The executive session is the part of the meeting designed to leave no written record, and an AI summary would create exactly the record the session exists to avoid.

Board Directors Using AI on the Pack: What the CEO Owes

Directors have started handing the pack to an AI tool the night before the meeting. Some use something the company approved. Plenty use whatever is already open on the laptop, which usually means a personal account on a consumer product whose terms of service were written for someone planning a long weekend in Lisbon.

I see this from both chairs. As CEO of Visiting Media I prepare the material, and on the board of ACT | The App Association I am one of the people reading what management sends. Reading is a part-time job with a full-time pile of PDFs attached, and I understand exactly why a director reaches for a summarizer at eleven at night.

The pages that rank for this question come from law firms and director associations, and they are good. They are also written almost entirely for directors. Nobody seems to have told the CEO what to do about it.

What the legal guides get right

Confidentiality comes first, and it is the easy part to agree on. A board pack carries forecasts, deal names, personnel matters and the occasional privileged memo from counsel. Pasting that into a consumer chatbot hands it to a vendor whose retention settings the director has probably never opened, and privilege is the kind of thing you can lose without noticing until opposing counsel points it out.

The guides converge on approved tools too. Give directors a company-controlled account or a board portal with an assistant built in, and say in writing that personal accounts are off limits for anything nonpublic. They add a records point I agree with: a director’s AI summary can be discoverable, so somebody has to decide how long those drafts live.

And every one of them says judgment stays with the director. True. Also roughly as useful as telling someone to drive carefully.

I would sign all of it. It still leaves the person who wrote the pack out of the conversation.

The summary is now the first reader

Board packs get long because management is nervous. Every caveat the CFO wanted goes in, along with the footnote from legal and the line that says the forecast assumes a renewal closes next quarter, and the pack grows a few pages every cycle to protect people from questions nobody was going to ask.

A summarizer reads that pack the way a hurried director does, only with more confidence. It keeps the headline number. It drops the condition attached to the number, especially when the condition lives in an appendix table eight pages away.

Long documents get the same treatment inside my own agent fleet, which is part of why one agent there does nothing except check facts against the source. The drafting model writes lovely prose. It is cheerful about losing a qualifier.

The director then shows up having absorbed “bookings beat plan” and having missed “because two deals were pulled forward from next quarter,” and the meeting spends twenty minutes on a celebration that should have been a question.

Run the pack through a model before you send it

This is the step I would add, and I have not found it in any of the guides.

Before the pack goes to the portal, someone on management’s side runs it through the kind of tool directors are likely to use, with a plain prompt asking what a director most needs to know. Then a person compares that summary against the short list of things management most needs the board to understand. Whatever fell out gets rewritten. I borrowed the habit from how I test agents, where a frozen set of real past tasks catches what a model silently stops doing (the full version is in change control for an AI agent fleet).

The fixes are mostly mechanical.

Put the condition in the same sentence as the number. “Bookings beat plan, and two of those deals were pulled forward from next quarter” survives summarization intact. A footnote does not.
Open each section with the bad news. Models lean on the first paragraph of a section, and so do tired directors.
Delete any appendix table that quietly contradicts the slide in front of it, or reconcile the two in a sentence on the slide itself, since the summarizer will pick one of them more or less at random and you will not get to choose which.
Write the decision you need as a sentence that starts with the word “Decision.”

A good CFO already writes an earnings script this way. The board pack now deserves the same care, because its first reader is very literal and never asks a follow-up.

If an agent drafted part of the pack, the disclosure line I recommend on the AI page in the board pack (which sections, and who checked every figure) matters twice over. A model may now be summarizing a model, and the person who checked the figures is the only human fingerprint in the chain.

Who should own the directors’ tool

Here I part ways with some of the board-portal vendors. Management should not supply the summary.

The tempting move for a CEO is to hand directors a ready-made AI briefing, or an assistant the company’s IT team configured to answer questions about the pack. It is convenient. It also puts management in charge of the lens the board uses to examine management, and an independent board exists largely to prevent that arrangement.

My split is simple. The company runs the tool through its security review, because the company carries the data risk. The board owns the configuration through the chair or the corporate secretary, and the setting that matters most is who can see a director’s questions. A director who asks the assistant what the CEO is leaving out should be certain the CEO will never read that question.

I hold this view with my self-interest pointed the wrong way. As a CEO I would love to know what my board asks its assistant. That is precisely why I should not be able to.

Executive session

Nobody should be summarizing the executive session with anything, because it is the one part of the meeting built to leave no paper behind, and a summary is paper with better formatting.

What the board’s own policy should say

Keep it to a page. Long policies go unread by the people they govern, which would be an odd fate for a document about reading.

Name the approved tools. Ban personal accounts for nonpublic material, with no exception for “just this once on the plane.” Set a retention period for director-generated summaries, and line it up with how the company already handles directors’ notes. Say that directors’ questions to the tool are private from management. Say nothing at all about which model is best, because that paragraph will be wrong by spring.

This belongs to the full board or its governance committee, which is one more reason I argued against standing up a separate AI committee. Bots that join the meeting itself are a different and messier problem, and I covered them in AI note-takers in board meetings. Directors who have already supervised one agent for a quarter tend to trust a summary less than everyone else in the room. That is the right instinct, and it is cheap to teach.

Before the next pack goes out

Run it through a summarizer yourself and read what comes back.

If the one sentence you most needed the board to absorb is missing, the director’s tool was going to miss it too, and you have a few days to fix the pack instead of a two-hour meeting arguing about the wrong number.

This article is part of the Executive Leadership cluster, focused on board governance and the operating discipline required to run AI systems responsibly at the executive level.

Your Directors Summarize the Board Pack With AI. Write It So the Summary Holds.