As Bermuda considers guardrails for artificial intelligence in financial services, catastrophe modeller Karen Clark & Co says AI could soon be updating the risk tools used by re/insurers and insurance-linked securities investors in days rather than months.

In a new white paper, the Boston-based company said AI was ushering cat models into a third generation after earlier statistical and physical models.

KCC argued that the technology is promising for so-called secondary perils, including severe convective storms, winter storms and wildfires. Severe convective storms, which include hail, tornadoes and straight-line winds, generated $61 billion in insured losses globally in 2025, the third-highest total on record for the peril.

The company said conventional statistical models have worked well for familiar risks like hurricanes and earthquakes, which can be described using a limited set of variables. But it said they were less effective for fast-changing and complex weather systems.

KCC has used high-resolution physical models to simulate those weather conditions in real time. It now says AI can build on that work by continuously processing new atmospheric and claims data, helping to identify patterns and update models faster.

“Applied correctly, AI is delivering accuracy and efficiency,” the paper said.

“Re/insurers can expect more frequent model updates that can be implemented in days — not months — and that deliver the most accurate and current data for pricing and underwriting decisions.”

The implications reach beyond pricing.

KCC said its severe-convective-storm model takes in more than 30 gigabytes of satellite, radar and weather data each day to create estimates of hail, tornado and wind impacts by 7am Eastern Time.

Insurers can use that information to assess claims and losses as an event unfolds, the company said — an application that could be relevant to Bermuda carriers, reinsurers and ILS managers after major US weather events.

KCC said the model has more than eight years of event estimates, alongside more than 100 terabytes of high-resolution atmospheric data and tens of billions of dollars of claims data. It said those data sets could be used to train AI systems to recognise complex storms, including derechos, widespread, long-lived, straight-line wind storms produced by a fast-moving cluster of severe thunderstorms.

An analysis found that in 2024, Bermuda insurers’ total exposure to major disasters such as hurricanes and earthquakes rose 7.5 per cent year-on-year to nearly $220 billion.

The KCC paper comes as the Bermuda Monetary Authority consults on a proposed guidance note on the responsible use of AI across the financial sector.

The BMA has identified underwriting, catastrophe modelling, reserving, portfolio management and risk transfer among the activities where AI could matter. The proposed guidance would not create a separate approval regime for AI, but would apply existing requirements on governance, risk management, cybersecurity, outsourcing and record-keeping.

Regulated firms will still be responsible for an outcome even when an AI system is supplied by a third party, the BMA has said.

In cat modelling, insurers often rely on external models, but KCC’s paper argues that firms whose models are “even a few years old” may already be behind current weather trends.

KCC founder Karen Clark has previously told The Royal Gazette that severe convective storms are often underestimated and cannot be assessed reliably with models designed for hurricane risk.