ACM Digital Library: https://dl.acm.org/doi/10.1145/3539609
We build SkillBot - a natural language processing (NLP)-based system to automatically interact with voice assistant apps and analyze the resulting conversations. SkillBot identifies risky child-directed voice apps that were certified and published. Our system also discovers a novel risk called confounding utterances that may cause a user to interact with a different app than intended. Our findings suggest that inappropriate content exists in certified child-directed apps and children are at risk of unintentionally invoking non-child-directed apps.