Is it right to use AI to identify children at risk of harm?