Adversarial AI in insurance: an overview

Excerpt

The rapid and dynamic pace of Artificial Intelligence (AI) and Machine Learning (ML) developments is revolutionizing the insurance sector, especially in property and casualty insurance. AI can handle the enormous amount of information collected on clients to exploit predictive analytics. At the same time, AI generates challenges for underwriters and claims handlers. AI can be the target of so-called “adversarial AI”, which causes AI and ML tools to misinterpret the information provided to them and to generate outputs (recommendations, predictions, decisions, categorizations) favourable to the attacker who can be a competitor, a customer or claimant, or simply a wrongdoer. The issue is made worse by the fact that adversarial attacks, although dangerous, are rarely seen. This letter aims to raise attention towards adversarial AI and its zero-label feature, which needs to be handled in an actuarially wise, data-sensitive way. …