These AI applications and services should follow the principles of prudence and precaution, should be adequately tested for accuracy and robustness, and should be with meaningful human control.