Our research team was included in an industrial project to design autonomous shuttles. For the first stages of the project, an escort was in the shuttle for legal reasons and to “ensure safety”: he can take control of the shuttle at any time. The prescription addressed to this escort was to intervene as little as possible. We analyzed the real activity and lived experiences of the escorts on four sites of experimentation of the shuttle in different European cities by different types of observations and interviews. Results show that respecting the prescription lead to a transport service with very little traffic. Most escorts have gradually adopted behaviors to improve the utility of the shuttle service, scarifying security value. Designing from a “total security” perspective can lead to useless services and prohibited behaviors. We argue that “acceptable security” model is ultimately more secure in the real world.