Autonomous driving systems can only be released to the public after developers have demonstrated their ability to achieve adequately high levels of safety. Today’s hands-off autonomous driving systems are largely built with deep learning algorithms that can be trained to make the right decision for nearly every driving situation. These systems, however, lack the detailed requirements, architecture, and validation that is already being used today in the development of safety-critical systems, such as the control systems of commercial airliners. Road and air testing is clearly an essential part of the development process, but the billions of miles of testing that will be necessary to validate the safety of autonomous driving systems and software is clearly impractical.