Should boards use AI note-takers in board meetings?

No. Kaykas-Wolff agrees with the law firms that advise keeping automated recording and transcription out of board and committee meetings, including executive session. A transcript is a different thing from minutes, and a recording held on a vendor's servers raises privilege and discovery problems that a human secretary does not.

How do AI note-takers end up in board meetings without approval?

Usually through a calendar default. A director uses a note-taking tool at their day job, leaves it set to join every meeting on their calendar, and the board meeting is on that calendar. Kaykas-Wolff treats the result as a shadow agent: automation with credentials and a schedule that nobody approved, running under the director's name.

What should a board policy on AI note-takers say?

Kaykas-Wolff recommends two lines. No automated recording or transcription in board or committee meetings unless the board votes to allow it for a specific meeting, and an annual confirmation, alongside the conflicts questionnaire, that nothing on a director's accounts will auto-join a board meeting. The chair then removes any bot that appears before the first agenda item.

Who is responsible for a transcript created by a director's AI note-taker?

The director whose account the tool runs on. That account holds the recording, the transcript, the summary, and often a searchable archive of every meeting it joined, all under the vendor's retention policy and one person's settings. Kaykas-Wolff argues the fix is to change the auto-join setting at the source rather than rely on directors remembering to remove the bot.

Does banning AI note-takers fix weak board minutes?

No. Kaykas-Wolff sees thin minutes as the more common risk, where nobody can reconstruct why a decision was made. He pairs the ban with a human secretary who records what was decided and why, dissent included, within a few days. An approved tool on company infrastructure can help tidy those notes, but a person decides what the record says.

AI Note-Takers in Board Meetings: The Bot Nobody Invited

I sit on the board of ACT | The App Association, and I run Visiting Media, where a fleet of specialized agents handles a good share of my own operating work. So I look at AI note-takers from two chairs. As a director, I want the minutes to be dull and correct. As an operator, I know what a meeting bot actually is: an agent with a login and a copy of everything said in the room.

That second view is missing from almost everything written on the subject.

What the law firms agree on

The pages that rank for this question are mostly written by lawyers, and they line up well. Minutes are the legal record of what a board decided, and a transcript is something else. All of them raise privilege, since a recording of counsel’s advice sitting on a vendor’s servers hands the other side an argument for waiver. All of them raise discovery, because a verbatim transcript captures the thinking out loud that good minutes leave out on purpose, and that is exactly what a plaintiff in a shareholder suit wants to read. They worry about accuracy (a director’s sarcasm turning up in a summary as a position) and about candor, since people talk differently when a machine is writing it all down.

One well-known firm post puts its recommendation in the headline: don’t do it. The others land close by with more qualifications, asking for consent from everyone present, a carefully chosen vendor, a human who reviews every output, and transcripts deleted on a schedule.

I agree with nearly all of it. Keep the bots out of board and committee sessions.

The advice assumes someone decides.

How the bot actually gets in

In practice it is rarely a decision. A director signs up for a note-taking tool at their day job, where it earns its keep, and leaves the default switched on: join every meeting on my calendar. The board meeting is on their calendar. At 9:02 a new tile appears in the grid with their name and the word “Notetaker” under it, and everyone is too polite, or too busy, to stop and ask about it.

An hour later a transcript and a tidy summary land in that director’s inbox. Depending on the tool, they may also land with everyone on the invite, which can include an outside advisor who dropped off after the first agenda item. The summary shows up before the corporate secretary has opened a draft of the minutes, and the approved minutes will follow weeks later, by which point most directors have already read the bot’s version and quietly started treating it as what happened.

It is an agent, so govern it like one

What I just described is a shadow agent, the same pattern I wrote about in shadow AI agents: automation with credentials and a schedule that nobody approved, wearing an employee’s name. Here the name belongs to a director. The habits I use on my own fleet carry over with very little translation.

Start with ownership. The owner of that bot is the director whose account it runs on, which makes that director personally responsible for where the board’s words went. Most directors have never thought about it that way. Their account also holds the data: the recording, the transcript, the summary, and in many tools a searchable archive of every meeting the account has ever joined. That archive sits under the vendor’s retention policy and one person’s privacy settings. Strange home for a privileged discussion about an acquisition.

Then turn it off at the source. Asking directors to remember to remove the bot each time fails within a quarter, because remembering always fails within a quarter. The auto-join setting has to change, or the tool has to be told to skip anything on the board’s calendar. I learned with my own agents that “we stopped using it” and “it stopped running” are two facts that can sit months apart, which is most of what retiring an AI agent is about.

Last, ask who else got the summary.

A policy short enough to remember

The board can adopt this in five minutes:

No automated recording or transcription in board or committee meetings. That includes executive session, and it includes tools running on a director’s own device. The board can lift the rule for a specific meeting by vote, and I cannot think of a meeting where I would.
An annual confirmation. When directors fill out the conflicts questionnaire each year, they also confirm that nothing on their accounts will auto-join a board meeting. It costs one checkbox and moves the setting from a habit to a representation.

Then the chair enforces it in the first minute. If a bot tile shows up, the chair removes it before the first agenda item and says why, out loud, without making anyone feel caught. Doing that once is usually enough.

Management’s own preparation is a different matter. A CEO rehearsing the board deck with the CFO can use whatever approved tools the company allows, because that conversation was never going to be minuted. Even there I would want the tool on a company account rather than someone’s personal one.

Executive session

Executive session is where the directors talk about the CEO, and I say that as a CEO who would very much like to keep it unrecorded.

Where I part ways with the lawyers

One place. The firms treat an overly complete record as the danger. In my experience the bigger risk runs the other way: minutes so thin that nobody, including the directors who were there, can reconstruct why a decision got made. If the minutes say “discussion ensued” under every agenda item, the board has a record problem the bot did not cause, and banning the bot leaves that problem exactly where it was.

So I would pair the ban with a better human process. A secretary who writes down what was decided and why, dissent included, within a few days. If that secretary uses an approved tool on company infrastructure to tidy their own notes afterward, fine. A person still decides what the record says. That same line is creeping into management reporting, where more of the AI page in the board pack gets drafted by agents every quarter, and the question of who actually wrote it is worth asking there too.

What to do before the next meeting

Have the corporate secretary add a single line to the calendar invite: no recording or transcription tools. The chair repeats it at the start. Then ask whichever director knows these tools best, often the technology board advisor if you have one, to spend ten minutes after the meeting walking the others through where the auto-join setting lives in the tools they use.

Then go after the bigger exposure. The board meeting is four or five hours a quarter. Management’s note-takers are running in customer calls and performance conversations every day, on a mix of company and personal accounts, and almost none of it is on a register. That belongs on the list of what boards should ask about AI, phrased as a count: how many meeting bots are running on company calls, and how many of them run on accounts the company does not control. If management cannot answer, the board has learned something useful about the rest of its AI governance, too.

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.