Insurance Pricing When Risks Are Artificially Generated: A Dynamic Control-Theoretic Framework for AI-Driven Hazards

Abstract

The proliferation of artificial intelligence systems across critical infrastructure has precipitated a fundamental transformation in the nature and structure of insurable risks. Unlike conventional hazards, which arise from exogenous stochastic processes amenable to historical frequency analysis, AI-driven risks exhibit endogenous dynamics wherein the risk-generating mechanism itself adapts, learns, and self-modifies in response to observed outcomes. This paper develops a rigorous mathematical framework for pricing insurance products when the underlying risk process is artificially generated through algorithmic decision-making systems. Through extensive Monte Carlo simulation, we demonstrate that traditional actuarial techniques systematically underestimate tail risk when applied to AI systems, with Value-at-Risk measures requiring upward adjustment of 24–36% and Conditional Tail Expectation increasing by 30–48%.