Trump Is Going to Raise Your Insurance Premiums
Private climate risk modelers are also increasingly leaning on large language models, or LLMs, in order to predict the behavior of storms and analyze risk, leading companies like Microsoft, Google, and Nvidia to roll out their own AI-branded forecasting models. While LLMs have helped spur some promising advances in predicting the weather, these products are still generally based on data generated by the public sector. If that data ceases to exist—or if it becomes outdated and unreliable due to cuts—then proprietary modeling could start to break down in ways that may not be immediately obvious to the customers who use those products, or to the researchers unable to assess how their results are tabulated. “AI is all well and good, but you still need good-quality inputs,” Condon told me. “AI can only be a supplement, not a replacement.”
Even before Trump 2.0, government researchers struggled to assess America’s vulnerability to extreme weather as temperatures rise. That’s partially because predictive modeling is based on relatively small historical datasets, and rising temperatures are changing how extreme weather behaves in unpredictable ways. Storms are slower and dump more rain, while wildfires burn hotter, faster, and at unusual times of year. In its 2023 report on extreme weather risk, the President’s Council of Advisors on Science and Technology urged that more coordination between NOAA, FEMA, and other federal agencies was needed to provide “more accurate and actionable information to guide decision-making and policy at all levels,” and remedy the “lack of high-quality estimates of extreme weather probabilities for most locations and types of events.” The council also explicitly noted the limitations of private-sector climate risk modeling, and the importance of interagency coordination for improving its reliability. “While a burgeoning industry is beginning to provide climate risk information, much of this is of questionable quality,” the report notes, “either because it has not been transparently skill-scored to show that it can predict past events, or because it relies on methods that have been shown by the academic literature to have significant bias.”
The stakes of accurate climate data are enormously high. As traditional mortgage lenders deny loans to prospective buyers in higher-risk areas, online “fintech” lenders have taken the opposite approach, offering financing on terms that may undercount the risk certain properties face from wildfires and other hazards. To offset their own risk, some lenders have also begun selling off their riskier mortgages, including in coastal areas, to Fannie Mae and Freddie Mac. A coming wave of climate-related foreclosures—which First Street predicts could soar 380 percent over the coming decade, and account for up to 30 percent of foreclosures—threatens not just housing but, potentially, the financial system more broadly.