Insurance companies are using generative AI to create more precise models of natural disasters, but this technology also has its limitations. Researchers warn that AI models can produce ‘hallucinations’ - events that look plausible but violate basic laws of physics. This raises concerns about the accuracy of these models and whether they will lead to better risk assessments or just more business for insurers. Why it matters: The increased use of generative AI in catastrophe modeling could lead to more insurers covering previously uninsurable regions, but may also reveal higher losses than previously thought, requiring larger capital buffers.