Should a sitting CEO serve on an outside board?

Kaykas-Wolff's view is yes, one seat, approved in writing by the CEO's own board, at a company that is not in a crisis and does not buy from the CEO's company. He counts nonprofit seats against the limit too, because their board packs take the same evenings to read.

What is the AI risk when a CEO joins an outside board?

If the CEO runs AI agents with access to their inbox, calendar or memory, the outside company's board material flows into systems the CEO's own company owns. An agent can store a confidential fact and surface it later in an unrelated draft, and it can also pull the CEO's own company data into comments meant for the other board.

How should a CEO keep AI agents away from outside board material?

Use a separate email address and portal login for the outside board that no agent can access, write that exclusion into each agent's access grants and check it in a quarterly access review, and put outside board meetings on the main calendar as untitled busy blocks so scheduling agents and note-takers have nothing to read or join.

Should a CEO tell the outside board they use AI agents?

Yes. Kaykas-Wolff recommends disclosing it to the chair or general counsel on joining and explaining how the agents are fenced off, and including the same paragraph in the approval memo to the CEO's own board.

Should a CEO Serve on an Outside Board? Fence Your AI Agents First

Search for whether a sitting CEO should take an outside board seat and the search firms and governance writers give you a cautious yes. Take one seat, or none if either company is in trouble. Pick it for what you will learn, and get your own board’s blessing before you accept.

I would sign most of that.

I became CEO of Visiting Media in November 2024, and I sit on the board of ACT | The App Association. I have been a trustee of American Conservatory Theater in San Francisco since 2017, and I hold advisory seats with HSMAI Las Vegas and the International Luxury Hotel Association. I also run a fleet of AI agents that triage, draft, summarize and remember things on my behalf, and that last detail changes the outside board question more than any of the guides admit.

What the standard advice gets right

The learning is real. Sitting on the other side of the table is the fastest way I know to become a better CEO for your own directors, because you feel firsthand what a bloated pre-read does to a Sunday evening, and you come home with a short list of things you will stop doing to your own board.

So is the time. A seat runs well past the meetings, and the expensive hours are the ones nobody schedules, like the call that comes when something breaks at the other company in the same week something breaks at yours.

Approval first. Tell your own board before you say yes, put it in writing, and mean it when you offer them the chance to say no. Then check for overlap with your customers and competitors, which the guides cover well.

One quibble. Brad Feld wrote in 2012 that every CEO should be an outside director of exactly one company, and he left nonprofit boards out of that count. I would put them back in. The A.C.T. board pack takes an evening to read, same as any other. If a seat sends you a pack, count it against your hours.

Your agents already read your mail

Nobody writing about outside board service seems to have noticed where the other company’s material ends up. It lands in the same places as your own work. The board pack link arrives in your inbox. The chair’s confidential note about the CFO shows up on a Tuesday night, wedged between a customer escalation and an invoice. If you run agents with read access to those surfaces, they read it too.

That is a confidentiality problem with your name on it.

The outside company’s information now sits in systems your company owns, under your company’s retention settings, reachable by whoever administers your agent stack (possibly someone on your own staff). The agent that turns my inbox into a morning brief has no idea that one message belongs to a different fiduciary duty. Neither does its memory. If the memory stores “the other company is weighing a sale,” that sentence can resurface three months later in a draft about something else entirely, addressed to someone who should never see it.

The leak runs the other way as well. An agent helping you prepare comments for the outside board will reach for whatever context it has, and what it has is your own company’s customer data, pipeline and plans. You owe both companies a wall, and the agent cannot see either one.

Nobody intends any of this. The plumbing does it.

A test you can run tonight

Search your agents’ memory for the other company’s name.

How to fence the seat

Give the board work its own identity. Use a separate email address for the outside board, and grant none of your agents access to it, or to the board portal. This is the fence that does most of the work, and it costs an afternoon.
Write the exclusion into the grants. Each agent’s access should name what it cannot touch, then get checked in the same quarterly sweep I described in AI agent access reviews, starting with the CEO’s own grants.
Strip the calendar. Outside board meetings go on your main calendar as untitled busy blocks, so the scheduling agent sees a conflict and the note-taker has nothing to join. AI note-takers in board meetings covers what happens when one gets in anyway.
If something already leaked, purge it and tell the chair.
Put the fence in your approval memo. When you ask your own board for permission to serve, one paragraph on how the agents are kept out of the other company’s material answers a question a careful director should be asking anyway, and it shows your board you thought about the risk to their company too.

Tell the other board, too

Disclose it to the chair or general counsel when you join. Boards are already writing policy about directors putting materials into AI tools, a problem I looked at from management’s side in directors summarizing the board pack with AI. A sitting CEO with an agent fleet is the director most likely to cause that kind of leak by accident.

That same director is also the one best placed to explain the controls in plain language, which is probably part of why the other board asked you to join. Describing how you fenced your own agents is a better first contribution than any slide about AI strategy.

A vote is a different job from advice

Some of what I do is advisory, and I covered that role in what a technology board advisor owes a CEO. An advisor can be wrong in private. A director votes, and carries duties that follow the information wherever it goes, including into a vector database you forgot was there. The fence matters more for a board seat than for an advisory one, and I would build it before the first pack arrives, while the board email folder is still empty.

Take the seat

One of them, approved in writing by your own board, at a company that is not in a crisis and does not buy from you.

The guides are right about all of that. They just assume the CEO is the only one reading the mail. If you have spent the last two years building agents that read everything so you don’t have to, the first piece of governance work on your new board is your own. For the wider set of rules that keep a fleet accountable, start with delegation of authority for AI agents.

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.

Should a CEO Serve on an Outside Board? Take One Seat, and Fence Your Agents Off From It