Due to the emergence of powerful Large Language Models (LLMs), many open-domain generative educational chatbots have evolved in the past few years, with many more to be expected as best practices and reliable methods emerge. In this paper, we consider a particular best practice that has not received much attention yet: the handling of student questions that are sensitive, politically inappropriate, or contain vulgar language. Although handling such questions wrong can have negative consequences and decrease the trust of users, parents, and teachers, our survey of papers introducing open-domain chatbots reveals that only a small percentage of papers consider them explicitly. In our survey (N = 55 papers), we make a case for why this problem should receive more attention, how it is currently being handled, and which steps future research should take. Moreover, our survey is, to the best of our knowledge, the most complete one considering only open-domain chatbots.