AI – unintended insurance impacts and lessons from “silent cyber”

AI has changed the way companies operate and will revolutionise global business practices further.¹ For example, generative AI that produces text, images, videos or other outputs, comes with a multitude of benefits, but also risks.² And, as the number of AI incidents rises rapidly,³ so too is C-suite level awareness of the dangers.⁴

AI is set to transform the insurance industry across various dimensions.⁵ Enhanced underwriting, AI-supported customer services, claims processing automation and predictive analytics for fraud detection, are just a few examples of the potential benefits. All told, however, the main utility of insurance remains risk transfer, in this case as protection against a variety of risks that their clients may face through their own increasing use of AI.

Harm caused by AI can be “material or immaterial, including physical, psychological, societal or economic”.⁶ It is unlikely that a single insurance policy will cover all potential risks that AI presents. As of today, AI risks are neither explicitly mentioned, limited nor excluded in policy language, and different exposures may be covered by different policies already in existence. Insurers need to analyse how AI-risks are dealt with in existing policies, and how to best deal with them in the future. Which risks are already covered, and which perhaps only silently so (in other words unintentionally, due to perhaps ambiguous language). It could be that the latter require alternative costing and risk assessment approaches to those inherent in existing policies, and/or even completely newly designed insurance products.

With silent cyber, the industry has already learned lessons. There have been instances where some risks have been covered in non-cyber line policies, even though that was not the intention. With silent AI, it is time to prevent repetition of the same mistakes by understanding which risks traditional policies already (silently) cover. With the fast development of AI and associated regulations,⁷ some of today’s assumptions may turn out to be wrong or incomplete. However, that should not prevent discussions around silent AI from starting now.

A scenario-based approach can help build understanding of AI risk use-cases. The scenarios and their impacts can be assessed following a structured process and set of questions, to understand how current policy wording/coverages would apply to specific risk cases. With these insights, insurers will be better placed to devise AI-risk transfer solutions that match customers’ future protection needs.