THE AI ESTATE RISKS THAT INSURANCE ACTUARIES ARE NOT MODELLING
A claim is denied. The adjuster cites the model. The model was trained on data from 2021. The underwriting environment has shifted materially since then. Nobody flagged it because nobody was watching for it. The insurer has a model risk policy. The insurer has an artificial intelligence (AI) governance framework. What the insurer does not have is any operational mechanism connecting those documents to that decision. When the complaint reaches the regulator, the policy is not a defence. It is evidence of awareness without action.
This is not a hypothetical. It is the structural condition of most insurance AI estates today. The gap between what an insurer’s algorithmic systems are doing and what its governance documentation says they are doing has a name. It is the AI estate gap, and it is now a primary regulatory exposure. The question for chief risk officers is not whether it exists in their organisation. It almost certainly does. The question is how much it will cost when enforcement makes it visible.
The problem is not non-compliance, it is false compliance
The most dangerous finding from cross-jurisdictional AI governance work is not that organisations are ignoring their obligations. Most are not. They have invested in policies, frameworks and documentation. The danger is that this investment has created a plausible appearance of control without the operational architecture to support it. Governance documentation exists. Continuous, auditable oversight does not. When a regulator asks for evidence of functioning controls rather than evidence of good intentions, the gap becomes immediately and expensively apparent.
Jul-Sep 2026 Issue
Kriterion Lab
Saint Mary’s University