In the absence of leadership, all efforts to govern will fail. In the absence of governance, all efforts to lead will fail.
To govern is a conscious, deliberate act of leadership. It is not a process, and it will always have instruments of measurement, observation, audit and protocol attached to it, correctly, because leadership exercised blind is not leadership anyone should want. But the sum of those instruments is less than the whole they were built to serve, in the same way a ship’s log, however completely kept, is not the hand that held the wheel.
A useful pair, corrected from an earlier draft of this argument, makes the point cleanly. A ship’s captain holds an accountability that cannot be delegated further down the chain, for the vessel, and for every person aboard it who cannot leave and cannot appeal to anyone beyond the ship. A ship’s surgeon exercises clinical judgement within that same enclosed responsibility, treating what needs treating, deciding what needs deciding, answerable in turn to the captain for how that judgement was exercised. Neither role is the instruments each of them uses. A chart is not a captain. A surgeon’s instruments are not a surgeon. The instruments inform; the office that holds the accountability is a different kind of thing entirely, and no accumulation of better instruments changes what kind of thing it is.
This is not an abstract point, and it is not only a historical one. The identical error is being made right now, in real time, about a technology genuinely new enough that the mistake has not yet had twenty five years to harden into settled vocabulary the way Clinical Governance did. Machine-level controls, measures, gauges and monitors that describe what an AI system, or an autonomous agent acting on an organisation’s behalf, is actually doing are being called AI governance. They are not. They are control, monitoring and audit, precisely the same instruments this whole account has already named, wearing a newer label. And the boardroom’s own use of AI is not, by itself, governing with AI either. A tool’s presence in the room does not upgrade a board’s process into an act of governance. It remains an instrument on the table until judgement has actually been exercised on what it shows.
This carries a real legal weight, not only a rhetorical one. Directors who fail to adopt proper system level controls for the AI their organisation deploys risk being found wanting in their ordinary duty of care and skill. But adopting those controls is a floor, not a ceiling. It does not discharge a board’s responsibility to govern what the controls report, and it does not remove accountability for the pattern those instruments should have revealed over time. It is not, and should never be argued to be, accountability for every single outcome regardless of whether it was foreseeable; that claim overreaches in exactly the way this whole account has been careful not to. But a recurring pattern is a different matter from an isolated event, and a board’s instruments exist precisely to make a pattern visible before it becomes a headline. Missing what a pattern was already showing is a failure of governance in a way that one unforeseeable event, properly controlled for, is not.
Twenty five years ago, Bristol built the right instrument in response to a real and terrible gap between what was known and what was acted on. The instrument was never the failure. The failure, then and now, is mistaking the instrument for the leadership it was built to serve, whether the instrument is an audit committee or a machine. Governance was never the systems. It was always, and remains, the board’s own act of leadership, exercised in full view of everything the instruments can show it, and answerable, in the end, for what was done with that sight.
