Artificial intelligence is introducing a new type of risk that could threaten underwriting and risk modelling across the insurance industry, according to Edin Imsirovic, AM Best director, in an interview with Best’s Review.

Mr Imsirovic said: “AI model poisoning is a hidden risk threatening insurance underwriting.”

The term refers to a form of data manipulation in which attackers subtly corrupt AI training data, leading to distorted outputs or flawed risk assessments.

“For an industry that relies heavily on complex third-party data, the potential implications — from financial losses to reputational damage — may be profound,” he said.

The warning follows AM Best’s analysis of how insurers are using artificial intelligence to streamline operations, detect fraud and enhance underwriting.

Mr Imsirovic noted that even small-scale data manipulation can alter how models behave. “A compromised model might continue operating undetected, quickly producing errors that accumulate over time,” he said.

The issue is emerging as Bermuda’s insurers use more and more advanced modelling tools.

At the same time, cyber and AI risks have entered the top-ten list of global business threats, according to Aon’s 2025 Global Risk Management Survey. The Bermuda Monetary Authority has also called for greater operational resilience among regulated entities.

Analysts say that as AI becomes more embedded in underwriting and catastrophe modelling, maintaining control over data quality will be key. AM Best said it would continue monitoring how insurers safeguarded AI systems against emerging threats.