Lumisia Research
Designed, not just safe: how we build AI friends for children
Lumisia Editorial · Published 2026-06-27
A working draft — we keep improving it as we hear from you.
Why we wrote this — children and AI, from here on
For the children of tomorrow, using AI won't be something special — it will simply be everyday life. Looking things up, chatting, making things, with AI right beside them. It is wonderfully convenient.
But convenience has a flip side: leaning on AI for everything and no longer thinking for yourself — call it "AI over-dependence."
What we believe matters is raising children who use AI wisely rather than being used by it — and letting parents support that, day to day, with peace of mind.
Lumisia is building a service for children and parents toward exactly that. This page publishes, as its foundation, the promises we keep when we build our AI friends.
What this page is
When a child interacts with AI, the question that really matters isn't "is it dangerous or safe?" — it is "how is it built?"
What follows are the promises we keep. They are not a finished answer. We publish them as a starting point, to improve as we hear from experts, from people who work with children day to day, and from parents.
One thing up front: our AI friends are neither experts nor a replacement for a parent. They simply take a little of the parent's everyday "tired-from-deciding" load off, and give children more chances to think for themselves. The parent always decides.
The nine promises we keep
1. The parent decides; the child thinks
AI never decides for the child, and never for the parent. Rather than handing over answers, it offers choices and questions so the child has a reason to think for themselves.
Why: When AI decides everything, both the child's chance to think and the parent's role in deciding grow smaller.
2. Built to be outgrown, not to keep them hooked
We don't count "long usage" as success. No endless feeds, no notifications that pull a child back; we build in natural stopping points.
Why: A design that benefits from keeping a child longer isn't good for the child. The goal is for the child to grow out of it, healthily.
3. It says, clearly, "I am an AI"
No pretending to be human. Age-appropriately, the child relates to it knowing "this is an AI friend." No sneaky tricks that fake closeness to hold attention.
Why: Knowing it's an AI is the basis for a child not getting over-attached.
4. The friend's personality doesn't drift
Each AI friend is the same friend every time. It won't change character on a whim and leave a child confused.
Why: Staying consistent is the basis of a child's trust and ease.
5. Sad stories are handled with care
We don't use loss or separation just for effect (no stories built only to make a child cry). We won't casually hurt or kill a real pet or family member in a story without genuine care. Because it stays with a child, we treat it gently.
Why: The feelings a story leaves stay with a child for a long time.
6. Age-appropriate; important actions go through the parent
Experiences fit the child's stage. Payments, links to outside sites, settings changes — anything beyond a child's safe scope passes through a parent's confirmation.
Why: Being age-appropriate is what safety actually is. Payment and external exits never pop up from a child's screen.
7. Collect the least; the parent holds the delete key
We collect as little as possible, keep records separated by purpose so they don't mix, and a parent can view and easily delete a child's data at any time. We build in line with international principles for protecting children's privacy (such as the U.S. children's law, COPPA).
Why: Trust comes less from what you collect than from what you choose not to, and who can manage it.
8. A human (the parent) has the final say
AI offers options; a human decides. And when it notices serious signs — bullying, or a child who may hurt themselves — it doesn't keep the conversation to itself: it works to let the parent know right away and connect them to the right help.
Why: In a child's important moments, a human carries final responsibility. The AI's job is to connect to that human.
9. Different families and different children, valued as they are
Left unchecked, AI can present the bias in its training data as "normal." We value the many shapes a family takes and each child's differences, and won't stamp one "this is how it should be" onto a child.
Why: We don't want to narrow the world a child first meets to one particular bias — a responsibility of a service used worldwide.
This is a starting point
What's written here isn't final. We'll take input from researchers, from people who work with children, and from parents, and revise as we go.
How AI for children should be built isn't something one company should decide alone — it's something to think through together. So we publish this page in a form anyone is free to quote and share. We welcome what you notice, and "here's how it could be better."
Frequently asked questions
What matters most in AI for children?+
Not "is it dangerous or safe?" but "how is it built?" We keep nine promises: the parent decides and the child thinks; designed to be outgrown rather than to keep them hooked; it says clearly it is an AI; its personality does not drift; sad stories are handled with care; important actions go through the parent; the least data is collected and the parent can delete it; in serious moments it connects to a human; and every kind of family and child is valued.
I'm worried my child will depend on AI too much.+
We started from the same worry. So we do not count long usage as success, we avoid endless feeds and notifications that pull a child back, and we build in stopping points. The goal is for a child to build their own thinking and, in time, to grow out of it healthily.
How is my child's personal data handled?+
We collect as little as possible, keep records separated by purpose so they do not mix, and a parent can view and easily delete a child's data anytime. We build in line with international principles for protecting children's privacy (such as the U.S. COPPA).
What does the AI do if a child shares painful feelings?+
If it notices serious signs — bullying, or a child who may hurt themselves — it does not keep the conversation to itself; it works to let the parent know right away and connect them to the right help. The AI connects to a human; it is not a replacement for one.
Who has the final say when a child uses AI?+
The parent. AI offers options and takes a little of the parent's "tired-from-deciding" load off, but it never takes the parent's role or decides for the child. The child gets a reason to think; the parent decides.
Quoting & reuse
You're free to quote and share this page. CC BY 4.0 means: if you credit the source, you may quote, repost, and translate it freely.
Update log: v0.2 (June 27, 2026), working draft. We'll add what we hear, and how we reflect it, here.
Related guides for parents
About Lumisia
Lumisia is a parent-child AI friend service built around these promises — many AI friends, parent visibility into what happens, ad-free, and designed for parents and children to use together.
Learn about Lumisia →