We develop a tractable framework to study the optimal design of stress scenarios. A principal wants to manage the unknown risk exposures of a set of agents. She asks the agents to report their losses under hypothetical scenarios before mandating actions to mitigate the exposures. We show how to apply a Kalman filter to solve the learning problem and we characterize the scenario design as a function of the risk environment, the principal’s preferences, and the available remedial actions. We apply our results to banking stress tests. We show how the principal learns from estimated losses under different scenarios and across different banks. Optimal capital requirements are set to cover losses under an adverse scenario while targeted interventions depend on the covariance between residual exposure uncertainty and physical risks.