AI safety for kids: the questions worth asking

Not 'is AI good or bad?' Kids smell that sermon. Ask: Who can see what I type? What happens if it is wrong? When do I need a person? How do I stay kind when the model is not a person and cannot be hurt — but other humans can?

WonderMind does not offer open child-to-child messaging. That is a safety design, not a missing feature. A PIN is not an email. A first name is not a school record. If another product asks for a birthday, school, and open DMs, that is a different risk profile even if the mascot is cute.

Teach kids to reward 'I do not know' from a model. Teach them not to jailbreak for cruelty. Teach them that crisis feelings go to a trusted adult. The Guide is instructed to drop the lesson there. A demo mailbox is not 24/7 counseling — say that out loud.

Sit nearby for Sprout and Explorer. Preview stage language if it feels off. Export and deletion are in-product on Free. If a district is asking for a DPA, send them /dpa and do not pretend the draft is signed.