Dealing with edge cases is the ultimate limit to achieving safe operation of embodied or physical AI systems. Some edge cases are truly novel objects or events. However, many problematic edge cases involve special contexts changing the typical reaction to known objects and situations. For any AI system working in an open-world environment, there will always be another edge case that machine learning has not been trained on. The good news is that remote assistants can provide a way to manage edge cases, if done thoughtfully. Remote assistants play a role in safety while fulfilling multiple different roles. AI-based systems need to ensure safety upon detection of an edge case while giving remote assistants the time and situational awareness required to help recover from an encounter with an edge case safely.