Most senior executives reading this have already built the AI accountability structures the moment seemed to demand. The accountable officer was named. The AI register is current. A board paper was written, and the committee charter was updated to bring AI into scope. None of that was wrong, and none of it was easy. It was the right work for the pace at which decisions moved when the policy was drafted.
In Brief
- The AI accountability structures most organisations have in place were designed for a decision cadence AI no longer operates at.
- Only 21% of organisations report mature governance for agentic AI, and responsible-AI maturity averages 2.3 out of 5 (Deloitte, 2026; McKinsey, 2026).
- AI accountability is now a top-three leadership imperative for 2026, and Australian boards rank AI as their number one concern (LHH, 2026; KPMG Australia, 2026).
- Naming an officer, maintaining a register, and updating a committee charter satisfies an audit question, not a design question.
- The board-ready answer is an independent structural assessment of whether the accountability structures in place are real or paper.
That pace has since changed. AI systems are now making, or materially shaping, decisions inside the organisation that previously required human deliberation, sign-off, and an audit trail. They do so at a cadence the accountability mechanisms in place were never designed to clear. The question those mechanisms were built to answer is no longer the question the board will ask.
A different question is forming
The question forming, in boardrooms and audit committees across the corporate and public sectors, is structural. It is not whether someone is accountable for AI. That has been assigned. The real question is whether the accountability mechanisms in place — the officer and register and committee and policy — were designed for the decisions AI is actually making in the organisation today.
Deloitte’s 2026 survey of 3,235 leaders across 24 countries found that only 21% of organisations report mature governance for agentic AI, with the remainder lacking clear decision boundaries, real-time monitoring, or audit trails (Deloitte, 2026). McKinsey’s 2026 research, drawn from around 500 organisations, found average responsible-AI maturity at 2.3 out of 5, with only about one in three at maturity level 3 or above on strategy, governance, and agentic AI governance (McKinsey, 2026). LHH’s 2026 C-Suite research, spanning 2,530 companies, identified AI accountability as one of the three core leadership imperatives for the year (LHH, 2026). KPMG’s January 2026 release of its survey of 274 C-suite and board executives put AI-related concerns in the top three challenge positions, with 63% ranking new technologies as their number one worry (KPMG Australia, 2026).
Those numbers do not describe a leadership failure. They describe a design gap. Accountability has been assigned at the level of named roles and documented policies. The design of those roles and policies — what they cover, the cadence they operate at, the evidence they produce, who they answer to and when — was drawn for the pace at which decisions moved at the time. That pace has moved on.
What follows is a pattern that capable, governance-literate executives reliably reproduce. The accountable officer is named, but no independent body has tested whether that officer’s mandate covers the decisions AI is actually making, or at what frequency. The AI register is current, but the inventory was built against the use cases known at the time it was written, not the agentic patterns operating now. The committee charter cites AI, yet the meeting cadence and approval thresholds were calibrated to a slower flow than the organisation now produces. Each artefact looks defensible on its own. The gap appears once they are tested as a system against the decisions AI is actually making.
The accountability design was never updated
This pattern persists for reasons that have nothing to do with the calibre of the executives carrying it. The standard governance toolkit — the named officer, the register, the policy, the committee — was built for a world in which decisions of consequence moved at meeting cadence. Quarterly papers. Monthly committees. Annual reviews. The instruments were designed to clear that cadence. AI does not operate at it. Increasingly, AI does not move through those instruments at all. The accountability has been assigned to the right people. The design of the mechanism through which they discharge it was never updated.
That distinction changes what the board paper needs to say. A board told “we have an accountable AI officer, an AI register, and an AI policy” is being told the right thing by the metric of two years ago. A board asking whether the organisation’s AI governance is designed for the decisions AI is making in 2026 is asking a different question, one that no single named officer, however senior or capable, can answer from inside the structures they inherited.
The liability builds between board meetings
If the design gap is not independently assessed, what follows is not a single failure point. It is a slow accumulation of accountability the executive carries but cannot evidence. Each AI decision made between board cycles — each agentic action and each model-driven recommendation that becomes a determination — accrues to the named officer’s role under the policy as written. Across a year, the volume of decisions for which accountability has been documented but not structurally tested becomes substantial. By the time one of those decisions surfaces as a question, whether from a customer, a regulator, or an internal audit, the board’s first question will not be “who is accountable?” That has been answered on paper. The first question will be “show me the evidence that the structure you put in place was designed for this decision.”
Most organisations cannot currently answer that question. The executives responsible are not uncertain about their accountability — no independent body has assessed the structures against the decisions AI is now making. An officer cannot grade their own mandate. A committee cannot test its own cadence. A policy cannot check its own coverage. Each of those acts requires an independent assessment of the design.
Where the senior executive stands now
The executive most exposed to this is the one whose organisation has done the visible work well: officer named, register maintained, policy current, board briefed. That work is not undone by the structural question. It is the platform on which the structural question can be asked. The executive whose organisation has done none of the visible work has an obvious problem. The executive whose organisation has done all of it has a less obvious but more dangerous one, because the documentation gives the appearance that the design question has already been answered when it has not yet been asked.
The board-ready answer is not a stronger officer or a refreshed policy. It is an independent structural assessment of whether the accountability structures in place are real or paper. That assessment is a different act from compliance review and a different act from policy refresh. It tests the design of the governance instruments against the decisions AI is actually making, and reports what the structure was built for, what it now faces, and where the two no longer line up.
The executive who has that assessment in hand before the board asks the question is operating with a different kind of confidence than the executive who has the documentation but not the design check. The documentation answers the question that was asked when the policy was written. The design check answers the question the board will ask next.
References
- Deloitte. (2026). Agentic AI is scaling faster than guardrails. Deloitte Insights.
- KPMG Australia. (2026, January). All things AI the biggest concern for Australian business leaders in 2026. KPMG Australia.
- LHH. (2026). LHH 2026 C-Suite research. LHH.
- McKinsey & Company. (2026). State of AI trust in 2026: Shifting to the agentic era. McKinsey & Company.
