In recent years, the application of artificial intelligence algorithms has accelerated the development of autonomous driving. However, their stochastic and black-box characteristics bring uncertainties to autonomous driving, i.e., they may result in unexpected behaviors when confronting unknown unsafe scenarios. How to monitor the health status of onboard AI algorithms and protect them safely in real-time is one of the key challenges in SOTIF research. This report focuses on the self-surveillance & protection of onboard AI algorithm risk, discarding the assumption that AI algorithm always performs perfectly, systematically considering the risk of upstream and downstream algorithms for autonomous driving system, and exploring real-time surveillance & protection technologies for the AI-based perception-prediction-decision-making algorithms.