How risk modelers like Fathom and Verisk are using AI and diffusion models to bypass the limits of physics-based “cat” models to predict natural disasters
Catastrophe scientists are pushing past the limits of physics-based models, improving how insurers calculate riskLinkedIn:Alberto CanalandAndrew SmithLinkedIn:Alberto Canal:Great piece in the Financial Times today that looks at how AI is transforming natural disaster prediction — adding to the traditional physics-based models …Andrew Smith:One of the questions I'm asked most often is: “How are you using AI?” — The Financial Times spoke with Oliver Wing and have done …
Context & Ripple Effects
Related coverage shows AI entering disaster risk work through real-time flood monitoring, wildfire forecasting and data-driven parametric insurance. This report extends that progression into the core catastrophe-modeling systems insurers use to assess exposure.
The significance is not simply faster forecasting: Fathom and Verisk are applying AI techniques where conventional physics-based catastrophe models have limits, potentially changing the inputs behind insurance risk decisions.
First-order effects
- Fathom and Verisk can add AI and diffusion-model outputs to catastrophe-risk assessments, giving insurers another modeling approach alongside physics-based systems.
- Insurers using these vendors may revise how they evaluate natural-disaster exposure, subject to validating the new models against established methods.
Second-order effects
- More usable AI catastrophe models would pressure rival model providers and specialist analytics firms to demonstrate comparable predictive performance, transparency and reliability.
- Changes in modeled risk can flow into underwriting, pricing, coverage terms and the design of data-driven products such as parametric insurance, especially in perils already served by AI monitoring tools.
Third-order effects
- If insurers gain confidence in AI-led catastrophe modeling, model competition may shift from primarily physics-based simulation toward hybrid systems that combine physical understanding with learned patterns.
- That shift would raise the importance of model validation and governance: AI models that influence insurance capacity and affordability will face greater scrutiny over how they behave in rare or changing disaster conditions.
The trend: This is part of the broader move from AI as a disaster-monitoring aid toward AI as an input to the financial infrastructure that prices and transfers climate-related risk.