Start Here
The board's AI responsibility has two halves. Most boards govern only one. What this publication argues, and where to begin.
Over the past two years, most boards have done the visible work of AI governance. They have approved investments, adopted policies, added AI to the risk register, commissioned management reporting, and, in many cases, assigned oversight to a committee. Directors have attended briefings and can now speak with reasonable fluency about models, use cases, and regulation. By the standards of how quickly governance usually adapts to new technology, this is fast work, and boards deserve credit for it.
All of it, however, points in one direction: outward, at the enterprise — the systems the organisation builds and buys, and how management deploys them. That is half of the board’s AI responsibility. The other half points inward, at the board itself: how directors use AI in their own work — preparing for meetings, interrogating papers, forming views, and reaching decisions. This half is largely ungoverned. Most boards have no position on it, and many directors are already using these tools privately, without guidance, in ways that touch confidential information and shape judgements the whole board then owns. The exposure is specific: confidentiality obligations breached by pasting board papers into consumer tools, collective decisions quietly influenced by unexamined outputs, and accountability that cannot be delegated resting on reasoning nobody minuted.
This publication covers both halves, and it is one of the few places that takes the second seriously. The method is consistent. Each edition takes one question a board will actually face, names the mechanism behind it — the cost, the risk, the accountability, the decision — and ends with what a director should do differently. It is not news coverage, not vendor content, and not reassurance.
A few convictions run through the archive, and they are a reasonable map of where to begin.
AI decisions are board decisions. The consequences — strategic, financial, legal, reputational — sit squarely within the director’s remit and cannot be delegated to the technology function. This has been the publication’s position since the first edition, and everything else here builds on it.
Directors’ own use of AI needs an order of operations. Form your view first, then use AI adversarially to stress-test it. Reverse the order and the tool shapes the judgement rather than sharpening it — and the board inherits a view no one actually formed.
Principles are not controls. A board that adopts AI principles for its own use has changed nothing until some of those principles are treated as controls, with owners, tests, and consequences. Edition 18 — Principles Are Not Controls — makes this argument and carries a downloadable guide, Principles for Responsible AI Use by Directors, written to be taken into a board meeting.
Advantage does not live in the model. The same AI is available to every competitor at roughly the same price, so whatever advantage exists must come from what the organisation owns around it. Edition 16 — AI Has No Moat. Your Organisation Might. — sets out the questions boards should be asking about that.
Connected decisions governed in separate rooms create risk. A board that approves AI investment in one committee and climate commitments in another has made a single decision in two rooms, and the gap between those rooms is where the exposure sits. Edition 17 — AI’s Hidden Environmental Footprint Is Becoming a Board-Level Risk — applies this to reporting that investors will soon read.
If you are new here, start with those three editions and take the guide into your next board conversation. The archive holds the rest — automation levels, productivity and organisational design, how AI changes the way directors interrogate information — and each piece stands alone.
I’m Karim Harbott, a Chartered Director, board adviser, and executive educator; the About page has the longer version. The publication is free, the archive is open, and every edition arrives by email if you subscribe.