Understanding AI risk

Purpose The rapid diffusion of artificial intelligence (AI) has elevated AI from a specialized technology to a digital infrastructure. This paper develops a structured framework for understanding AI risks arising from an insurance and risk-management perspective, with a particular focus on its distinction from cyber risk and the resulting implications for insurability. Design/methodology/approach The paper presents insights from the academic and policy literature on AI, cyber, and operational risk and combines them with first empirical evidence from AI incident and litigation databases. Given the scarcity of historical loss data, the analysis relies on conceptual classification, comparative risk analysis and scenario-based reasoning, drawing parallels to the development of the early cyber insurance markets. Findings AI risk is partially overlapping, yet distinct from cyber risk. While cyber risk is mainly driven by external attacks and system compromise, AI risk is rooted in internal system behavior, including model errors, biased outputs, autonomous decision-making and governance failures. AI risk exhibits strong accumulation potential, complex attribution, and extreme tail risk, limiting the applicability of traditional actuarial methods. Some AI-related risks (e.g. operational failures with clear causality, AI-related liability claims and AI-enabled cyber incidents) seem insurable, whereas systemic, correlated and attribution-opaque risks remain challenging and potentially uninsurable. In general, AI shifts risks from idiosyncratic to systemic and from exogenous to endogenous, which are difficult developments from an insurability point of view. AI-related incidents also generate measurable negative abnormal stock returns; capital market instruments such as insurance-linked securities may complement traditional risk transfer as loss data and incident taxonomies mature. Practical implications For insurers, the results underline the need for narrow coverage definitions, conservative limits, scenario-based pricing and enhanced underwriting standards focused on AI governance and control. Policymakers can improve insurability by clarifying liability regimes and transparency. Lessons from the early cyber insurance market (e.g. importance of minimum-security standards) offer guidance for the gradual development of AI insurance products. Originality/value The paper offers a classification of AI risk, systematically contrasts it with cyber risk, and links AI risk characteristics to core insurability criteria. By integrating risk classification, incident data and insurance economics, it provides a structured foundation for future empirical research and AI insurance product development.