SafeConnect AI: A Formative Evaluation of Child Online Safety, Parental Views and Privacy Across 45 Family Dyads

This study presents a formative evaluation of SafeConnect AI, a prototype exploring explainable, age-sensitive and privacy-conscious approaches to child online safety. The research involved 45 parent–child dyads (90 participants) across seven countries, including children aged 7–16. Parents evaluated simulated safety alerts, explanation preferences, graduated interventions, parental visibility options and privacy considerations. Children completed age-adapted baseline questions and interacted with fictional, age-appropriate online safety scenarios. The findings indicate that 73.3% of parents considered real-time alerts useful, 68.9% preferred alerts accompanied by explanations, and 66.7% favoured graduated interventions. In addition, 64.4% expressed an intention to consider adopting the proposed approach. Children's responses provided further insights into online safety awareness, warning comprehension, help-seeking confidence and perceptions of autonomy. The study highlights design considerations for developing responsible, human-centred child online safety technologies that balance protection, transparency, parental involvement and children's privacy. SafeConnect AI was evaluated through simulated, rule-based interactions. It was not integrated with real messaging platforms, and the study does not establish the technical accuracy or real-world effectiveness of an operational AI detection system. The findings should therefore be interpreted as formative evidence to inform future research and participatory design. Keywords: Child online safety; Responsible AI; Human-centred AI; Child privacy; Parental mediation; Explainable AI; Digital wellbeing; Human-computer interaction; Formative evaluation.