Boards are quietly signing away something they have no authority to give. As autonomous AI agents move from pilot to production, directors are approving systems that make and execute decisions at machine speed,
Boards are quietly signing away something they have no authority to give. As autonomous AI agents move from pilot to production, directors are approving systems that make and execute decisions at machine speed, and in doing so they are extending the board\'s decision-making perimeter to a system that carries none of the board\'s accountability. The World Economic Forum put the principle plainly this year: you can outsource execution to a synthetic system, but not fiduciary duty (WEF, 2026). When an agent acts, it acts inside the board\'s liability, whether or not the board understood what it had authorised.
This is the governance question of the decade, and most boards are answering it by accident. They are voting, often blindly, on where human judgement ends and machine authority begins, and recording the vote nowhere.
The readiness data is sobering. Deloitte finds that while the overwhelming majority of organisations intend to adopt agentic AI within two years, only around a fifth have a mature governance model for it (Deloitte, 2026). The distance between those two numbers is unowned liability, sitting quietly on the board\'s books until the day it is called.
In my research across 6,000 executive leaders, this is the sharpest form of what I call synthetic authority: action without appointment or accountability. A human director with an agent\'s reach over outcomes would be the most scrutinised person in the institution. The agent is scrutinised by almost no one, precisely because it appears in no governance framework as an actor at all. The board still holds the fiduciary duty. It has simply lost sight of where that duty is now being discharged.
A chair for a high-end consumer goods company based in Paris asked me a question last year that has stayed with me. An agentic system had been proposed for a part of the business that touched customers directly, and she said, simply, \"If this thing does something we would never have sanctioned, who is liable, and how would I even know it had happened?\" No one in the room could answer the second half. That inability, the not knowing, is the real exposure. A board can defend a decision it made and got wrong. It cannot defend a decision it never knew was being made in its name.
The instinct is to reach for technical controls and audit trails. Necessary, but insufficient, because the failure here is not technical. It is a failure to define the boundary of delegation before the delegation happens, and boundaries drawn after an incident are called findings rather than governance.
My recommendation is singular. Draw the delegation line explicitly, and put it in the minutes. For every material decision an autonomous system is permitted to make, the board should record what objective it serves and which named human owns the outcome. If leadership cannot trace an agent\'s decision back to defined intent and a named human, the system should not be operating. Fiduciary duty is not delegable. The only real choice a board has is whether it discharges that duty deliberately or discovers it, far more expensively, in a courtroom.
For Board Advisory, Strategic Consulting, Keynote Speaking or Private Executive Advisory enquiries, please write with a short note on the decision, audience or transformation context. Suitable engagements are shaped through an initial private conversation.