The CEO Is Now the Chief AI Officer. Who Is Checking the CEO?
Edition 21 — Why boards should welcome CEO ownership of AI and still treat it as a delegation to be documented, bounded, and independently assured.
For most of the past two decades, technology was the one consequential subject a chief executive could safely leave to someone else. Strategy, capital, talent, and reputation sat with the CEO; technology sat with the CIO, and the arrangement suited everyone until it failed. The failures were expensive and repetitive. Transformation programmes were approved as technology projects and starved of the business ownership they needed to change anything. Cyber risk was filed under IT until a breach turned it into a reputational and regulatory event. Technical debt accumulated quietly for years because nobody at the top of the house could price it, and boards discovered the bill only when the modernisation programme arrived asking for nine figures. Anyone who has led a large technology organisation will recognise the pattern: the decisions that determined competitive position were treated as plumbing, and the people accountable for competitive position stayed out of the plumbing.
Against that history, the headline finding of BCG’s AI Radar 2026 reads as progress. Seventy-two per cent of chief executives now describe themselves as their organisation’s main decision-maker on AI, double the share a year earlier, according to a survey of 2,360 executives, including 640 CEOs, across 16 markets. Accountability for the most consequential technology of the era has finally moved to the person who owns strategy and capital allocation, which is exactly where boards spent twenty years wishing technology accountability would sit. The instinct is right, and boards should say so before they say anything else. The trouble is that what has been built around the instinct is very little.
The delegation migrated. Nobody wrote it down.
“Main decision-maker” in the BCG survey is a self-description, and the governance question is whether the paperwork agrees with it. In most companies, the schedule of matters reserved for the board, and the delegation of authority beneath it, were drafted before generative AI existed and say nothing useful about it. The decision rights the survey describes were never formally allocated; they migrated towards the most engaged executive while the documents stayed still. That would matter less if the sums involved were still modest, but they are not. The same survey projects corporate AI spending to double from 0.8 per cent to roughly 1.7 per cent of revenue in 2026. For a FTSE 250 business turning over £2 billion, that is in the region of £34 million a year, and rising, which moves AI from a rounding error inside the technology budget to a capital-allocation question in its own right. A commitment of that size in M&A or capital expenditure would exist in writing, with thresholds, approval routes, and reporting obligations attached. For AI, in most organisations, it exists as a survey answer. The first board action this edition argues for is unglamorous: reconcile the chief executive’s de facto authority over AI with the documented delegation, and do it before approving the next budget line rather than after.
A conflict boards would name anywhere else
The same research reports that half of CEOs believe their job depends on getting AI right. Read that as a governance professional rather than as a technologist. A decision-maker whose personal survival is tied to a programme’s perceived success has a structural incentive to continue that programme, to enlarge it, and to report it favourably. Boards know this dynamic intimately from M&A, where the executive sponsoring a deal is understood to be the least reliable judge of it, and where the response is procedural rather than personal: independent valuation, non-executive scrutiny of synergy claims, and post-completion reviews that compare promises with outcomes. Nobody considers this insulting to the sponsor. It is simply how boards handle enthusiasm attached to career exposure.
Klarna is the case worth sitting with, precisely because it does not involve a bad actor. Sebastian Siemiatkowski championed the company’s AI-first service model personally, told OpenAI he wanted Klarna to be its favourite guinea pig, and presided over an assistant that the company said was doing the work of 700 agents while headcount fell by around a third. The strategy formed part of the story told to investors ahead of the company’s listing, and, by his own account, the company had over-rotated: hiring for human support reopened in 2025, and he has since described the episode plainly as having gone too far on costs at the expense of quality. Nothing in the record suggests dishonesty, and his candour in reversing is to his credit. What the record does not show is any mechanism, independent of the chief executive, that tested the flagship bet before the customers did. With AI, the identical conflict boards would police in a deal sponsor is being scored as a virtue, because a CEO who takes personal ownership of the technology looks like a leader rather than a sponsor. The practical consequence for boards is narrow: the metrics by which the AI programme is judged cannot be defined solely by the executive whose tenure rides on the programme looking successful.
The board’s teacher is the person it is meant to challenge
Concentration of decision rights is only half of the loop. The other half appears in BCG’s Split Decisions survey from May, which polled 625 leaders, among them 351 CEOs and 274 board members. Three-quarters of board members rate their own AI knowledge as adequate. Their chief executives are less convinced: more than half say hype is distorting boardroom judgement. BCG’s recommended remedy is for the CEO to personally lead an AI upskilling session for the board, and the remedy is the most revealing finding in the report. If the board’s understanding of AI arrives principally through the executive it is meant to oversee, then the board has no independent basis on which to challenge the strategy it is being asked to approve. This holds whatever the integrity of the individual CEO; it is a property of the structure, not a suspicion about the person. Boards long ago accepted the equivalent point in financial reporting, which is why the external auditor reports to the audit committee rather than through the finance director whose numbers are being audited. Nobody proposes that the finance director personally teach the audit committee everything it knows about accounting. On AI, that is roughly the arrangement now being recommended as good practice, and celebrated as chief executive engagement.
The boards pushing hardest are the least equipped to push
Honesty requires saying that boards are not the passive victims of this arrangement. The same Split Decisions research finds that around 60 per cent of CEOs believe their boards are rushing AI transformation, that 35 per cent think their boards overestimate what AI can actually replace, and that the board members least confident in their own AI knowledge are the most likely to insist the organisation is moving too slowly. That last finding deserves a moment of discomfort in every boardroom, because it describes urgency generated by uncertainty rather than by analysis. Pressure of that kind transfers risk to the executive without adding any challenge to the strategy. It also explains why chief executives have absorbed the decision rights so easily: a board that demands speed but cannot interrogate substance leaves a vacuum, and the CEO has filled it. The test a board should apply to itself is simple to state. If it cannot articulate the conditions under which it would slow or stop an AI investment, it is applying acceleration, not oversight, and the distinction will matter a great deal when one of these programmes fails at scale.
What the delegation should actually say
None of this argues for taking AI away from the chief executive, and boards that respond by pulling decisions upward will repeat the M&A-committee error of substituting their own enthusiasm for the executive’s. The argument is for putting the delegation on the same footing as every other delegation large enough to sink the company, and three controls do most of the work. First, reserved matters: name the AI decisions that stay with the board, including spend above a defined threshold, deployments that create regulatory exposure under the AI Act’s transparency obligations, which took effect on 2 August, and programmes that substitute for the workforce at scale. Second, an independent assurance line: internal audit or an external reviewer reporting on AI programme health directly to the audit or risk committee, without executive intermediation, exactly as boards already insist upon for cyber and for financial controls. Third, a challenge budget: a modest standing allocation for external AI counsel commissioned by the board itself, so that the board’s understanding of the technology has at least one source the chief executive does not control. A CEO who welcomes those three controls is demonstrating the accountability the surveys describe. A CEO who resists them has told the board something more useful than any survey could.
If this is the kind of analysis you want to arrive before the board pack, subscribe to AI in the Boardroom. I write for directors, executives, and advisers who need to govern AI with the same rigour they bring to capital, audit, and M&A, and who suspect that enthusiasm at the top of the house is no substitute for a written delegation. Future editions will keep working through the decisions boards cannot afford to hold as survey answers.