AI vendor spillovers and insurance capital: a network-based operational risk methodology

Ogunsanya, Adedayo, Ogunsanya, Feyi
Journal of Business Economics · Springer Nature · 2026-11-02 · DOI: 10.1007/s11573-026-01294-7

Abstract

Insurers increasingly rely on external artificial intelligence (AI) systems, shared model families, data providers and cloud infrastructure. This paper develops a network-based operational risk scenario framework that links these dependencies to annual loss distributions and internal capital analysis. It combines a failure taxonomy, dependency scores, frequency-severity scenarios, centrality, common-model dependence and control adjustments. For the constructed dependency network, the 500,000-year 99.5% value at risk (VaR) is $146.9 million for standalone vendor risk, $286.5 million with network amplification and $315.0 million under common-model dependence; stronger controls reduce it to $203.9 million. The network increment is $139.6 million (95.0%). Additional analyses examine propagation caps, bridge removal, fixed reference weights, equal edge selection, scale-matched Pareto severity and governance factors within one amplification rule. They show that propagation strength, event allocation and the centrality scale materially affect capital magnitude. The implementation amplifies the loss associated with a selected dependency; it does not simulate a cascade of additional failures at other nodes. The framework can be populated with insurer-specific data on dependencies, exposures, controls, and losses to support own risk and solvency assessment (ORSA), internal capital assessment, and vendor concentration analysis.

Read at publisher →

Similar articles