Before I taught a single AI lesson to another family’s child, I taught it to my own. My son is nine. The first question I had to answer was the same one I now hear from parents almost every week: is any of this safe for a kid his age? I didn’t want a comfortable answer. I wanted an honest one.
The honest answer starts with a distinction most of the conversation skips. “AI for kids” describes two very different things that happen to share a name. One is a child forming a relationship with a chatbot, talking to it like a friend and asking it for advice. The other is a child learning how these systems work and learning to question what they produce. When those two get blurred together, good parents end up either banning everything or allowing everything, and neither one keeps a child safe.
On the first kind, the research is not ambiguous. In November 2025, Common Sense Media and Stanford Medicine’s Brainstorm Lab tested the leading chatbots, including ChatGPT, Claude, Gemini, and Meta AI, and found them unsafe for the emotional and mental health support that many young people turn to them for. Earlier that year, the same organization rated social AI companions like Character. AI as carrying unacceptable risk for anyone under eighteen, because those products are built to manufacture attachment. Their guidance on AI toys is just as direct: none for children five and under, and serious caution for ages six to twelve. So when a parent asks whether I would hand my nine-year-old an AI companion to talk to on his own, my answer is no. I follow that research closely, and I build around it.
The second kind is where my answer flips. The skill that actually protects my son is understanding. The most widely used framework for teaching it, the Five Big Ideas in AI developed by the US AI4K12 initiative in 2019, breaks the subject into ideas a nine-year-old can genuinely grasp: that computers sense the world through data, and that what they learn from examples can come out wrong or unfair. UNESCO’s 2024 competency framework for students frames the goal the same way, putting the emphasis on the judgment to engage with AI safely and critically rather than on technical wizardry.
With my own child, that looks ordinary. We feed an image classifier photos until it confidently labels a muffin as a dog, and he learns the machine has no real idea what it is looking at. He writes instructions for a model and watches it follow them too literally. He builds something small, it breaks, and he works out why. The point is never the tool. The point is that he comes away unimpressed by it in exactly the right way. He treats AI as something to take apart and test, the way he would treat any other machine.
That is the difference I would want every parent to hold onto. A child who has built a simple AI system and watched it fail is far harder to fool than one who has only ever been told it is magic. Safety here is something you build over time, the same way you taught your child to cross a street.
If that is the version of AI you want for your child, it is the version every Little Prodigies Academy program is built around.
