When Do Children Delegate Inquiry to AI? Measuring Developmental Reliance

Developmental overreliance on AI is an emerging child-safety concern, but it remains unclear if and when consultation begins to displace children's own inquiry. We probe this boundary in two exploratory studies. In Study~1, 10 children aged 6--12 were descriptively more open to consulting AI for informational problems than for emotional or social-navigation problems, motivating a closer examination of AI as an epistemic resource. Study~2 embedded an AI-framed assistant in an established causal-learning task in developmental cognitive science. After observing the assistant successfully solve a demonstration problem, 21 children aged 6--10 repeatedly chose whether to explore independently, receive evidence from the assistant's investigation, or receive the assistant's full solution before exploring. Children chose to explore themselves first on 83.3\% of test rounds (participant-clustered 95\% CI [71.4\%, 94.0\%]). Despite non-ceiling performance (children only correctly identified the full causal structure on 58.3\% of rounds), they requested the assistant's exact answer on only 4.8\% of rounds, and every round beginning with AI assistance was followed by further child investigation. While these studies do not demonstrate overreliance or developmental harm due to AI, they reveal a behavioral distinction between asking AI for information and delegating active inquiry, and provide a low-risk proxy evaluation for studying when that boundary shifts across developmental and task contexts.