Is ChatGPT safe for kids?

Not really, and it isn't trying to be. ChatGPT is a general-purpose AI assistant, built primarily for adults, and OpenAI's own terms of use set the minimum age at 13 — anyone 13 to 17 needs a parent's permission. That single fact answers most of the question: a product with a 13+ age floor was not designed as a space for a seven- or ten-year-old to use alone.

That doesn't make ChatGPT sinister. It's a genuinely useful tool, used by hundreds of millions of adults every week, including a lot of parents. The issue isn't malice, it's fit. ChatGPT will answer almost anything and generate almost anything the moment you ask, because that's the job it was built to do. Nothing about it assumes a child is on the other end, or that every reply needs checking before it lands.

What OpenAI itself says

OpenAI has added real parental controls over the past year — you can link a teen's account to your own, set quiet hours, turn off certain content types, and get an alert if the system flags a serious safety concern like self-harm. That's a meaningful step, worth using if your teenager is on ChatGPT. But it doesn't let you read the conversation, it doesn't check every message before your kid sees it, and it isn't built for a child under 13 at all. The controls are patched onto an adult product for a supervised teenager, not built from the ground up for a young kid.

The deeper problem isn't safety filters — it's what the tool hands over

Set the age limit aside for a moment. The bigger issue for a kid's development is what ChatGPT does by default: it produces the finished thing. Ask it to write the essay, and it writes the essay. Ask it to draw the picture, and the picture appears. For an adult trying to get work done faster, that's the whole appeal. For a kid still forming the instinct to make things themselves, a tool whose default move is doing it for you teaches exactly the wrong reflex — reach for the answer, not for the struggle that builds the skill.

That's a different concern from "will my kid see something inappropriate," and it matters just as much. A tool can be perfectly safe and still be the wrong tool for a child's growth, if what it trains is dependence instead of capability.

So what does a genuinely kid-built AI space look like?

If ChatGPT is a general assistant with a 13+ floor and some bolted-on parental controls, what would the alternative actually require? Not a friendlier chatbot — a different architecture, built for a child from the first line of code:

• A walled garden, not the open internet. No strangers, no ads, nothing outside a bounded, kid-only space.
• Every message checked twice. Not "flagged if something goes wrong" — every single message between the kid and the AI reviewed by a second AI system before it ever reaches the child, automatically, every time.
• A parent who can actually see in. A weekly summary of what was discussed, an immediate alert if something needs attention, and the ability to review, freeze, or delete the account at any moment — not a settings page you have to dig for.
• An AI that asks questions instead of finishing the work. The point isn't just filtering out bad content; it's whether the AI hands your kid the answer or pushes them to find it themselves.

This is the actual design brief for an AI space built for a child. It's also, plainly, the description of what Kubrio built.

Kubrio's AI Crew: built kid-only, from the ground up

Inside every Kubrio app, kids work alongside the AI Crew — three kid-safe thinking partners named Krea (creative), Tek (maker), and Brio (questioner) — with one hard rule: ask a better question, never hand over the answer. The Crew will clear the drudgery (rendering a film, exporting a finished page) but never make the creative decision for your kid. Your child always knows they're talking to an AI, never a person. You can read the exact rule on the AI Crew.

On the safety side, Kubrio is a kid-only, ad-free, COPPA-compliant walled garden — no open internet, no strangers, no ads. Every message between a kid and the AI is checked by a second AI before it reaches them, automatically. Parents get a weekly summary of what was discussed plus an immediate alert if something needs attention, and can review, freeze, or delete the account at any time. Sharing is private by default; going public requires a parent's consent, and public pages show a first name only — no last name, no face, no contact information. Full detail lives on safety and safety and data.

That combination — walled garden, message checked twice, parent in the loop, and an AI that questions rather than answers — is what "safe for kids" actually requires. It's a different job than the one ChatGPT is doing, and neither is pretending otherwise.

Where this shows up for a real kid

In Discovery, a kid opens the always-available quest library and the Crew helps them pick a project and push through it — asking what they're trying to make, not making it for them. In Sketchling, a kid draws key frames of their own story by hand on real paper; the Crew bridges the motion between those drawings, and never draws the picture itself. In Book Club, a kid gets asked increasingly sharper questions about what they read, not handed a summary to skip the reading. Three different domains, the same rule underneath all of them: the Crew moves the kid forward, the kid's own hand and decisions are what's left at the end.

A checklist you can run on any AI tool

You don't need to take Kubrio's word for any of this. Before handing a kid any AI product — Kubrio, ChatGPT, or anything else — run it through four questions:

1. Is it a walled garden, or does it open onto the wider internet?
2. Is every message checked before your kid sees it, or only flagged after something goes wrong?
3. Do you get real parent controls — visibility, alerts, the ability to freeze or delete — or just a settings toggle?
4. Does it ask questions or give answers when your kid gets stuck?

An AI space built for a child should pass all four. A general-purpose assistant, however good, will fail at least the first two by design — which tells you exactly what it's for, and what it isn't.