Safety for autonomous driving (AD) is currently a hot topic both in industry and research. Especially for prototype vehicles the idea prevails that engaging a safety driver is sufficient to mitigate all safety concerns. We point out, why a safety driver in several cases is unable to cope with the situation, and needs additional support from the AD system. Especially at higher vehicle speeds (e.g. on highways), the reaction time of a safety driver is a pitfall. In this paper we analyze the situation, and propose a detection of challenging and limiting critical driving maneuvers, by using well known relations from vehicle dynamics. On the basis of this approach, we present the implementation into our autonomous driving demonstrator and show how to relax the response time requirements of the safety driver. Since highest safety requirements need to be met, the described methodology is implemented into a dependable embedded system. For our purposes an AURIXTM microprocessor with dedicated safety modules was chosen. Both, the hardware setup as well as the software implementation is discussed. Furthermore, the additional application of the proposed Safety-Gateway, to train safety drivers is reported too.