How should a voice AI system like Alexa handle potentially dangerous requests made by children, and what are the key design principles to ensure child safety?

1. Separate detection of speaker age and risk classification. 2. Use voice signals to detect age, while running a risk classification on the content of the request independently. 3. Ensure that framing does not influence risk assessment. 4. Combine both signals for a response strategy. 5. Process classifications in real-time and log high severity incidents.