The hard part of an AI phone assistant is not answering every unknown number; it is choosing the right boundary. A doctor s office, school, delivery driver, or new customer can look just as unfamiliar as spam. We are exploring controls such as contact and calendar context, repeat-caller signals, caller-intent questions, urgent-call escalation, and an immediate human takeover. What rule would make you trust an AI to answer: always screen first, answer only while you are busy, use different rules by caller type, or never handle certain categories? I am especially interested in the failure that would make you turn the feature off.