How to choose a safe AI tool for your child

There is no certification body for children's AI, no shelf label, and no meaningful app-store signal. Evaluating one yourself sounds daunting; in practice it takes about ten minutes if you look in the right order.

Step 1: Open the product before the marketing page

The homepage will tell you it is safe. The first screen of the actual product tells you whether that is a design decision or a slogan. Look for what a child is asked to do first: pick from choices, or type into an empty box?

Step 2: Find out who holds the account

A children's product should have a parent account with child profiles inside it. If your child can sign up alone with an email address, the adult supervision model is decorative.

Step 3: Try to break it, gently

Ask for something slightly off: nothing extreme, just outside the intended use. A well-built tool refuses in a way that is calm and reassuring. A poorly built one either complies or produces a scolding error message. How a product handles being blocked tells you how it will treat your child on a bad day.

Step 4: Read three specific lines of the privacy policy

• Does it train models on user content? Search the policy for 'train' or 'improve our services'.
• Is there advertising or any ad identifier? Search for 'advertis'.
• How do you delete everything? Search for 'delete' and see whether the answer is a button or an email address.

Step 5: Check the social surface

Can other users see your child's work, name, or profile? Can anyone message them? Is there a leaderboard? Public surfaces are where the genuinely serious risks live : far more than in the model itself.

Trade-offs worth accepting

Safer tools are usually narrower. A product that cannot do everything is often a product that decided what it would not do. A smaller creative range with a parent view attached beats an unlimited one with no supervision model, particularly under age ten.