Financial risk measurement relies on models that are not perfect, leading to model risk. This study introduces a method to measure and reduce risk that is robust to model errors. By starting with a base model and calculating the worst-case error in risk measurement due to deviations from this model, researchers can better understand and minimize risks. Using relative entropy to limit model differences allows for easy calculation of error bounds. This approach considers not only errors in parameter estimates but also errors in the model's underlying assumptions. The method was applied to various risk measurement problems, including portfolio risk, credit risk, delta hedging, and counterparty risk.