A Sandbox Approach to Regulating High-Risk Artificial Intelligence Applications