Only 24 Percent of Leaders Will Own the Decision. That Is the Real AI Risk.

The UN Global Dialogue on AI Governance concluded in Geneva on July 7, with all 193 member states present for the first time on this issue. The outcome was a shared statement of principles, formal agreement that the UN Scientific Panel on AI's preliminary assessment will serve as the technical reference for future negotiations, and a commitment to reconvene in New York in May 2027 (UN, July 2026).

That is a governance milestone. It is not a judgment milestone. A shared statement of principles tells 193 governments what they agree on. It does not tell a single executive what to do when an AI system recommends an action that is efficient, defensible on the data, and still wrong for the people who will live with it. This week's research answers a more uncomfortable question than whether AI is governed. It asks who is actually accountable when it is used. The data suggests fewer people than we assume.

Here is what is inside.

The Accountability Gap: Why only 24 percent of leaders say their executive committee is ultimately accountable for AI-informed decisions, and what the other 76 percent implies.

The Decision McKinsey Says Every Organization Must Make Now: What the State of Organizations 2026 study of 10,000 senior leaders identifies as the defining leadership requirement of this period.

Designing Who Decides: Why Deloitte's 2026 Human Capital Trends frames decision-making as a discipline to be designed, not a habit to be hoped for.

The Liability That Cannot Be Outsourced: What boards are learning about AI accountability the hard way.

THE ACCOUNTABILITY GAP
The Accountability Gap: Who Actually Owns The Decision

New research published this week found that 70 percent of leaders second-guess their own judgment when an AI system disagrees with them. Forty-six percent say they now rely on AI more than they rely on their colleagues. Sixty-five percent report that decision-making inside their organization has become less collaborative since AI adoption (Forbes, July 2026).

Read those numbers together and a pattern appears. Confidence in independent human judgment is eroding faster than AI's actual reliability is improving. Leaders are not deferring to AI because it has proven itself infallible. They are deferring because disagreeing with a system that processed more data than they can hold in their head feels increasingly difficult to justify in the room.

Here is the number that matters most. Only 24 percent of leaders say their CEO or executive committee is ultimately accountable for AI-informed decisions.

That is the actual risk. Not that AI makes a bad call. Systems make bad calls, and so do humans. The risk is a governance vacuum forming underneath the decision itself, where the analysis was AI-assisted, the recommendation felt authoritative, and no one in the room is prepared to say the decision was theirs.

Analysis processes information at a scale no leader can match. Judgment is what remains once the processing is done. It integrates context, values, relationships, history, and consequence into a decision someone is willing to be accountable for. When 76 percent of leaders are not confident that accountability sits clearly with anyone, the organization has not solved its judgment problem. It has outsourced the appearance of one.

MKINSEY INSIGHTS
The Decsion Every Organization Must Make Now

McKinsey surveyed 10,000 senior leaders across 15 countries for its State of Organizations 2026 research, built around identifying the decisions leaders need to make now to sustain performance through what the report calls an unprecedented confluence of geopolitical, technological, and economic disruption (McKinsey, 2026).

The finding that should reframe how leadership teams think about AI readiness: organizations led in a human-centric way report stronger trust, better decision-making, and greater resilience under disruption than organizations that lead primarily through process and metrics. Not despite operating in an AI-augmented environment. Because of it.

The logic holds up under scrutiny. When workforces increasingly comprise both AI systems and human employees working side by side, the leadership capability that differentiates is not who has adopted the most AI. It is who has built the organizational judgment to know where AI's contribution ends and human accountability begins, and who has made that boundary legible enough for an entire organization to operate inside it.

McKinsey frames this as a decision, not a trend leaders can wait out. Organizations that decide deliberately how judgment and analysis divide their labor are building a capability. Organizations that let the division happen by default are accumulating exposure.

THE DESIGNER VS DECIDER
Designing who Decides: Why Deloitte Calls This a Discipline

Deloitte's 2026 Global Human Capital Trends report makes a claim worth sitting with. It argues that organizations should treat decision-making itself as a strategic discipline, one that requires intentionally designing how humans and AI share judgment and accountability, rather than allowing that division to emerge informally through daily use (Deloitte, 2026).

Most organizations have not done this. AI tools were adopted for efficiency. Nobody convened a working session to decide which categories of decision an AI system may inform, which it may not touch, and who is accountable for the outcome in each case. The division of judgment labor has been left to whichever employee is closest to the decision at the moment it needs to be made.

That is how the accountability gap identified above actually forms. It is not a single dramatic failure. It is thousands of undocumented, ungoverned micro-decisions about how much to trust an AI recommendation, compounding across an organization until no one can say with confidence where human judgment was actually applied.

Deloitte's recommendation is specific. Design the decision architecture the way an organization designs any other operational discipline: with defined categories, explicit accountability, and a documented boundary between what is delegated and what is reserved. The organizations doing this now are treating it as a leadership function. The ones deferring it are treating a governance question as an IT rollout.

THE LIABILITY GAP
The Liability That Cannot Be Outsourced

Board-level research this year has converged on an uncomfortable finding for directors. Only 35 percent of boards report having integrated AI meaningfully into their own oversight activities, even as the systems they are meant to oversee scale rapidly across the organizations they govern (EY, 2026; Harvard Law School Corporate Governance, 2026).

The finding beneath that statistic matters more than the number itself. Boards continue to treat AI risk the way they treat other operational risk: as something that can be transferred through a vendor contract, an insurance policy, or a third-party audit. Legal and governance analysis this year has been blunt about why that framing fails. Infrastructure can be outsourced. Liability cannot. When an AI-informed decision made using a third-party system causes harm, accountability remains anchored to the institution that used it, regardless of who built or hosted the underlying model.

This is the same accountability gap identified in the executive-level research above, playing out one governance level higher. If 76 percent of leaders are not confident that accountability for AI-informed decisions sits clearly with anyone at the executive level, and only 35 percent of boards have built the oversight capacity to catch that gap, the exposure compounds rather than cancels.

The boards getting this right are not the ones with the most sophisticated AI oversight technology. They are the ones that have accepted, structurally, that no amount of governance architecture removes the requirement for a specific, named, accountable human judgment at the point of consequential decisions.

SHARE YOUR THOUGHTS
We value your feedback!

Your thoughts and opinions help us improve our newsletter. Please take a moment to let us know what you think.