The rapid adoption of artificial intelligence (AI) techniques is being used to develop increasingly capable autonomous vehicles. While the major focus has been on improving the performance and accuracy of AI techniques applied to autonomous vehicles, development towards functional safety has been lagging behind. This paper proposes Sentinel, a faulttolerant safety architecture, designed to mitigate safety concerns surrounding AI techniques employed by upcoming SAE J3016 Level 5 autonomous vehicles. The architecture draws inspiration from existing autonomous vehicle architectures as well as architectures in the related domains of AI and organic computing. An assurance case was constructed to demonstrate that Sentinel provides high level features that support compliance with SAE J3016 Level 5 autonomy and that Sentinel meets or exceeds the safety of other autonomous vehicle architectures.