Every profession that deploys consequential technology eventually faces the same structural problem. The people with the technical knowledge to implement governance are not the same people who bear accountability for it. Every mature regulated sector closed the gap the same way: by creating a role that combines both.
AI governance has the inversion without the solution.
The engineer who writes the system prompt, the governance policy, or the evaluation criteria has the technical power to make it structurally sound or structurally incomplete. They can see whether the scope boundary is well-formed, whether the acceptance criteria will hold, whether the document will still be valid in six months. They have no personal liability for any of it.
The executive or compliance officer who approves AI deployment has nominal accountability. They often lack the technical depth to evaluate what has been implemented. The EU AI Act places human oversight obligations on providers and sets fines up to €35 million. Both target the entity. Not the individual. Not the person who wrote the document.
Knowledge without liability. Liability without knowledge.
The actuarial profession created the appointed actuary. Pharmaceuticals created the qualified person. Aviation created the designated engineering representative. Nuclear safety created the safety case author. Each role exists because the inversion was costing lives or money at scale, and the profession decided the gap had to be closed. In every case the solution was the same: one named individual, technical depth, personal liability, and an external standard that management cannot override.
In your organisation, who is the appointed actuary for your AI governance documents? Who has the technical knowledge and the personal liability? If you cannot name that person, you have the inversion.