Future safety-critical systems will be highly automated or even autonomous and they will dynamically cooperate with other systems as part of a comprehensive ecosystem. This together with increasing utilization of artificial intelligence introduces uncertainties on different levels, which detriment the application of established safety engineering methods and standards. These uncertainties might be tackled by making systems safety-aware and enabling them to manage themselves accordingly. This paper introduces a corresponding conceptual dynamic safety management framework incorporating monitoring facilities and runtime safety-models to create safety-awareness. Based on this, planning and execution of safe system optimizations can be carried out by means of self-adaptation. We illustrate our approach by applying it for the dynamic safety assurance of a single car.