SimSimi has been a huge success because of its advantage that anyone can 'teach' it a conversation. However, because of these advantages, SimSimi has been most exposed to toxic language, a risk factor that no conversational AI has fundamentally addressed so far. With a focus on safety over fun, SimSimi Inc. created a universal and intuitive content policy and planned to have people examine every conversation scenario SimSimi has “learned” to date (approximately 140 million). This seemingly impossible plan is bearing great results after many years of hard work. For each sentence that SimSimi uses as an answer, votes with guaranteed sincerity are collected from more than 10 users of that language to build high-quality data on the toxicity determination of conversations. The data constructed so far amounts to tens of millions, and the neural network model trained on this data has shown high performance, enabling effective control of toxicity in multiple languages. This neural network model, called DBSC, is recognized as a solution necessary to secure safety in various fields related to online comments, games, broadcasting, dating, and social network services.