The Challenge
There’s a quiet shift happening in boardrooms that nobody has quite named yet. Directors are busy people, board packs are long, and AI is very good at summarising long documents. So a growing number of directors are doing the sensible-seeming thing: feeding the pack into a model, reading the summary, and walking into the meeting across the top of it. It feels like a productivity win, the same kind everyone is chasing everywhere else. But of all the places to quietly hand your reading over to a machine, the boardroom is the one where it carries a risk most directors haven’t thought through, because reading the papers is not an administrative chore that sits around the duty. In a real sense, it is the duty.
A director carries a personal duty of care and diligence. The detail varies by jurisdiction and you should take your own advice on the specifics, but the principle is settled and old: a director must bring an informed, independent mind to the decisions of the board. They cannot simply rely on what others tell them the papers say. They are expected to engage with the material themselves, to understand it well enough to question it, and to be capable of recognising when something in it doesn’t add up. The duty is personal and it is non-delegable, which is a careful way of saying you cannot get someone, or something, else to do the understanding for you and still claim to have discharged it.
The real issue
This is where AI summarisation runs into a problem that has nothing to do with how good the summary is. Suppose the model produces a flawless three-paragraph precis of a forty-page pack. The director reads it, feels informed, and votes. Have they brought their own mind to the material, or have they brought their mind to a machine’s account of the material? Those are not the same act, and the difference is exactly the thing the duty is concerned with. A summary, however good, is a set of choices about what matters, made by something that doesn’t know what this particular director would have found significant, doesn’t know the history behind the numbers, and doesn’t carry any of the responsibility for getting it wrong. The whole point of a director reading the pack is that a specific, accountable human applies judgement to it. Outsource that and you haven’t sped the duty up. You’ve skipped it.
It gets sharper with the things that matter most. The risk buried on page thirty-one, the inconsistency between two divisions’ figures, the footnote that quietly changes the meaning of the headline, these are precisely the things a summary is most likely to smooth over, because a summary is built to convey the gist and the gist is where danger hides. A director reading the actual pack might catch the thing that doesn’t sit right. A director reading the summary will catch only what the model decided to surface, and will have no way of knowing what it left out, because the nature of a summary is that the omissions are invisible. You can’t notice what you weren’t shown.
None of this means AI has no place anywhere near a board pack. It has a very valuable place, and this is the distinction worth holding onto, because it’s where the entire question turns. The useful place for AI is on the other side of the table, with the executives who write the papers, not the directors who read them.
When Ai works in board paper
When an executive uses AI to make a board paper clearer, tighter, better structured, easier to follow, that is straightforwardly good practice. A paper that buries its key point on page nineteen is failing the board; a paper that states it plainly up front is serving them. AI is genuinely useful for this. It can spot where an argument is muddled, where the structure fights the reader, where jargon has crept in, where the ask isn’t clear. Used this way, AI raises the quality of what’s presented to the board, which means directors are reading better papers and are better placed to do their job. The executive is using the tool to serve the board’s comprehension.
The director using AI to summarise is doing the opposite thing with the same technology. They’re using it to avoid the comprehension the role requires. One side uses AI to make the material more worthy of a director’s full attention. The other uses it to pay the material less. Same tool, and the line between good governance and a quietly abdicated duty runs precisely between them.
The need for a shared approach
The practical upshot for any board is not a ban, which wouldn’t hold anyway, but a shared understanding of where the line sits. Executives should be encouraged to use AI to make papers as clear and readable as they can possibly be, because that genuinely helps. Directors should understand that the readability is the gift, and that the gift is wasted, and the duty unmet, if the well-made paper then gets fed to a model so the director can avoid reading the thing that was carefully made readable for them. The better the executives get at writing clear papers with AI’s help, the less excuse there is for directors not to read them with their own eyes.
There’s a version of the near future where boards run entirely on machine summaries of machine-polished papers, where AI writes to AI and the humans engage with neither, and everyone feels efficient right up until something goes badly wrong and the question gets asked: who actually read this? It’s worth deciding now, deliberately, that the answer will be a person. The reading was never the overhead. The reading was the job.
This piece raises a governance question rather than offering legal advice. Directors and boards should take their own professional advice on how their duties apply to the use of AI in board processes.