In The Actuary, the official magazine of the Institute and Faculty of Actuaries, in the UK, a short article.
AI is often discussed as a tool for insurers: a way to improve pricing, underwriting, fraud detection, claims management and customer service. But it is also becoming an object of insurance. Firms use AI systems in customer service, legal drafting, software development, marketing, finance, logistics, medical triage, compliance, credit decisions and operational workflows. These systems can generate losses due to erroneous advice, discriminatory decisions, privacy breaches, intellectual property disputes, cyber incidents, automated pricing errors, contractual liability, professional negligence, reputational harm or failures of autonomous agents.
It is true that AI creates a new risk, but only partly. Many of the problems it raises are familiar to insurers: limited historical data; fast evolving technology; opaque models. The insured may know more than the insurer, and service quality may be difficult to observe. Failures may be discovered only after deployment, and claims may involve complex questions of causation and responsibility. These are not new actuarial problems. Catastrophe insurance, cyber insurance, professional liability, product liability, and technology errors and omissions already involve uncertainty, model risk, asymmetric information and delayed discovery.
What is distinctive is the scale and structure of dependence. AI risk is not just a risk attached to one firm using one tool: many firms may rely on the same model, provider, cloud infrastructure, application programming interface (API), dataset, plugin or agent framework, so a single failure can produce many losses at once. This is the core accumulation problem. The insurer is not only underwriting the insured firm – it is indirectly underwriting part of the technological stack on which that firm depends.
This connects AI risk to cyber risk, but with an additional layer. Cyber insurance already struggles with common vulnerabilities, shared software, cloud concentration and network effects. AI adds common models and common behaviours. A faulty model update, compromised API, contaminated dataset, systematic retrieval failure or recurring hallucination pattern may affect many insureds simultaneously. The loss process is not independent across policyholders because policyholders share parts of the same AI supply chain.
OpenEdition suggests that you cite this post as follows:
Arthur Charpentier (June 25, 2026). The price of proof: insurance policies for an AI-enabled world. Freakonometrics. Retrieved July 15, 2026 from https://doi.org/10.58079/16gsk