The question your board will ask

Boards govern through questions. A director does not run the systems, read the code, or check the outputs; a director asks management to demonstrate that the institution is under control, and judges the answer. This is the whole mechanism of oversight, and it is why the questions a board asks are not incidental — they are the instrument. When a new category of risk emerges, the governance system adapts not by having directors learn the technology but by adding a question to the standing list, a question management must be able to answer credibly and on demand. Agentic decision-making is now producing exactly such a question, and it is worth stating precisely, because the wrong version of it wastes the board's authority and the right version concentrates it on the one thing that matters.

The questions that miss

The tempting questions are the technical ones. Is the model accurate? Is it biased? Is it explainable? These sound like the right things to ask, and they are not wrong to ask, but they are the wrong questions for a board to own, for two reasons. First, they are unanswerable at the board's altitude: a director cannot evaluate a claim about model accuracy any more than they can audit the code, so the question collapses into "management says it is fine," which is not oversight but its appearance. Second, and more importantly, they are the wrong risk. Accuracy and bias are engineering and ethics problems that the institution should absolutely manage — but they are problems the institution manages for itself, on its own terms, on its own schedule. They do not, by themselves, generate the exposure that boards exist to guard against, which is the exposure that arrives from outside and cannot be managed on the institution's schedule.

The question that maps onto the board's actual duty is different. It is not "are our systems making good decisions?" — a question about performance. It is "if someone with standing demands that we account for a decision our systems made, can we?" — a question about answerability. The distinction is the whole point. A board's fiduciary concern is not that the institution occasionally errs; every institution errs. The board's concern is that the institution might find itself unable to answer for what it did — exposed, on someone else's timeline, to a demand it cannot meet, with the directors themselves in the frame for having failed to ensure the institution could. Answerability, not accuracy, is the governance question, because answerability is where the uncontrollable, outside-driven, board-level risk lives.

Directors are not obligated to guarantee that every automated decision is right. They are obligated to ensure the institution can account for the decisions it made — and that is a question with a yes-or-no answer.

Why this is a duty, not a preference

The reason the question is coming — the reason it will appear on board agendas whether or not any individual director pushes for it — is that oversight duty has a well-worn structure, and automated decision-making walks straight into it. The duty to oversee has never required directors to prevent every bad outcome; it has required them to ensure that the institution has systems in place to know what it is doing and to answer for it, and to make a good-faith effort to see that those systems function. When a domain of activity becomes large and consequential, the failure that attaches to the board is not the underlying harm — it is the absence of any system by which the institution could have monitored and accounted for that activity. Directors get into trouble not for the bad decision but for the missing apparatus that would have caught, or at least explained, the bad decision.

Automated decisions at scale are precisely the kind of large, consequential activity this duty reaches. Once an institution is making consequential determinations through systems, faster than humans can review them, the board is on notice that a domain of significant activity exists that it must ensure is answerable-for. And here the parallel to established oversight domains is exact: just as directors are expected to ensure the institution can account for its financial reporting and its compliance obligations — not by personally verifying every entry, but by ensuring an apparatus exists that makes accounting possible — they will be expected to ensure the institution can account for the consequential decisions its systems make. The apparatus is the point. A board cannot discharge the duty by trusting that the record could be reconstructed if needed; the duty is to ensure the apparatus that produces the record exists before it is needed.

What a good answer looks like

Because the question has a yes-or-no character, so does the answer, and directors should be unsatisfied by anything that dodges the binary. The inadequate answer is a description of how good the systems are — how accurate, how carefully built, how well-intentioned. That answers a different question. The adequate answer demonstrates the apparatus: for the consequential decisions our systems make, we capture, at the moment of decision, a faithful account of what evidence was consulted, what rules were in force, where the inputs came from, and enough preserved state to show the decision again; that account is kept in a form we cannot silently alter and an outsider can examine; and we have tested that it holds up when read by someone who does not trust us. A board hearing that answer can credit it, because it is a claim about a system's existence and properties, not a claim about a system's quality — and existence and properties are things a board can require to be demonstrated and independently checked.

The tell of a weak answer is that it points to logs. Management that responds to "can we account for our decisions?" with "we log everything" has misunderstood the question, and a sharp director will notice, because logging is capture for insiders and the question is about answerability to outsiders. The follow-up that exposes the gap is simple: "Have we ever taken a real, contested decision and reconstructed it end-to-end as if defending it to a regulator or a court — and did it hold?" If the honest answer is no, the institution does not know whether it can account for its decisions; it only assumes it can. And an assumption is exactly what the oversight duty forbids a board from accepting in place of a demonstrated system.

Why the board is the center of gravity

All of this is why the board — and the executive who must face it — is the true center of gravity of this market, more than any technical buyer. The technical owner evaluates the decision-making system and optimizes its performance. But the person who has to stand in front of the board and answer the answerability question owns a different problem, one that no amount of model performance solves, and one that is personally theirs: the institution's capacity to account for itself. That person does not need to be persuaded that automated decisions matter; they already lie awake over the question they will be asked. What they need is a credible way to answer it in the affirmative — a demonstrable apparatus, independently kept, that turns "we assume we could reconstruct it" into "we have, and it holds." The board will ask the question because its duty compels it. The only variable is whether the institution has built the answer before the question is put — because the one thing worse than being asked is being asked and having to say you do not know.

— Dispatches · Summit Cognitive