Artificial intelligence raises a “deeper question” around the ways more precise risk assessment will change the nature of insurance, Australian Prudential Regulation Authority member Suzanne Smith says.

AI could improve underwriting and pricing, but greater segmentation may make some risks harder or more expensive to insure, with implications for the breadth, accessibility and affordability of insurance pools, she told the Insurance Council of Australia Annual Conference on Friday.

“The challenge for insurers will be to capture the benefits of better risk information while understanding these broader consequences.

“It also raises questions about the boundaries of insurability and the respective roles of insurers, regulators, governments and communities in reducing underlying risk.”

Ms Smith said board governance must keep pace with AI innovation as characteristics making the technology powerful also risk more consequential failures.

“Errors can spread at speed and scale, amplifying cyber, data, model and operational risks, while creating new third-party dependencies beyond an insurer’s direct control.

“Opaque models can make outcomes harder to explain or reproduce when challenged, just as expectations for clear and transparent decisions are rising.”

Ms Smith said risks can arise through third-party providers, or through customers and others using AI, while the technology has cost implications, could weaken brand loyalty and may heighten anti-selection risk.

“The answer is not to hold back innovation. It is to ensure governance keeps pace. Boards do not need to understand every technical detail, but they do need a clear view of where AI is used, its material risks and outcomes, and who is accountable.”

Ms Smith says new technologies are amplifying interconnected risks and operational resilience should be more than a “compliance tick-a-box”.

Without strong governance and risk assessment “an entity that’s operating really well and on track can find itself derailed by an event that wasn’t on the radar”.

See Analysis.