Safety review

Is AI Safe for Kids? The Honest Answer

Published August 19, 2026 · By the Kidgeni team

There's no single answer, because 'AI' isn't a single thing — but the categories sort cleanly. AI companion chatbots: no — safety researchers call them unacceptable for anyone under 18, and regulators and courts spent 2025–2026 agreeing. General AI assistants like ChatGPT and Gemini: only with an adult alongside, for kids under 13. Creation-first AI tools — the kind where the output is a drawing or a story, not a conversation — are the lowest-risk shape, when they're moderated and a grown-up owns the account. The one universally wrong answer is not deciding: 81% of 9-to-12-year-olds already use AI, and more than 4 in 10 kids say no parent has ever talked with them about it.

Start with the real baseline

Common Sense Media's 2026 census — the first annual study of AI use by tweens and teens — found 86% of kids 9–17 use AI, including 81% of 9-to-12-year-olds (17% daily). Mostly for fun, then schoolwork. So the practical question isn't whether your kid will touch AI; it's which kinds, on whose terms. The same study's most striking number: more than 4 in 10 kids say no parent has ever talked with them about AI safety. Sorting the categories below is that conversation's prep work.

Category 1: AI companions — the documented no

Companion chatbots — apps engineered to be a friend, confidant, or romantic partner — are the category with a body of evidence behind the warning. Common Sense Media, working with Stanford Medicine's Brainstorm Lab, concluded in April 2025 that social AI companions pose unacceptable risks to anyone under 18. The record since: Character.AI banned open-ended chat for under-18s entirely (announced October 2025) amid wrongful-death lawsuits it and Google moved to settle in January 2026; the FTC opened a formal inquiry into seven companies' companion chatbots and minors (September 2025); California's SB 243 became the first state law regulating companion bots (effective January 2026); and a federal ban for minors — the GUARD Act — has cleared two Senate committees, though it is not law yet. When the company, the researchers, the regulator, and the legislature all point the same direction, parents can skip the ambiguity: not for kids.

Category 2: general assistants — with an adult

ChatGPT, Gemini, and their peers are built as adult tools with youth protections added — and that's exactly how the reviewers score them. Common Sense Media rates general assistants as moderate risk (a 'powerful creativity booster' among other things) while rating Google's kid and teen Gemini experiences High Risk for being adult models with filters bolted on. The vendors are moving: OpenAI shipped parental controls in September 2025 and launched a separate ChatGPT for Teens experience in August 2026; Google allows under-13s only inside Family Link supervision. The research caution that remains is emotional, not just content-based: a large OpenAI/MIT Media Lab study found heavier chatbot use correlating with more loneliness and dependence. The working rule from reviewers: ages 6–12 only with an adult present, and no minor using a chatbot for companionship or emotional support.

Category 3: creation-first tools — the lower-risk shape

The safest architecture isn't a better filter — it's a different loop. Tools where a kid makes something (art, a story, a book, music) and the session ends with an output don't run the mechanism that makes companions dangerous: the persistent relationship. Policy analysts have started saying this plainly — technology for children 'should act like technology, not like a substitute for human relationships' — and Australia's eSafety Commissioner draws the same line between assistant/creative chatbots and relationship-designed ones. The honest caveat: creation tools are lower-risk, not no-risk. The questions that remain are exactly the ones our safety checklist covers — is generation moderated, is the work private by default, who owns the account, and is the child's data used for training (which, under the 2025 COPPA amendments, now requires separate parental consent).

What the pediatric establishment says

The American Psychological Association's 2025 health advisory on AI and adolescents makes five recommendations parents can borrow directly: AI experiences should match a child's developmental stage; defaults should be protective; health-flavored answers need loud disclaimers; kids need actual AI literacy, not just rules; and their data deserves special protection. Notice what's absent: 'keep kids away from AI entirely.' The expert consensus is match the tool to the child — which is the whole argument of this page.

Our product-by-product reviews

The category rules above are the map; these are the street views. Each review is researched fresh, dated, and sourced:

Five rules that survive every category

The account belongs to the adult; the kid gets a profile — if an app wants your child's email, that's your answer.
No AI 'friends': anything designed to be talked to like a person is an over-18 product, full stop.
Prefer tools where the session ends with a thing — a picture, a story, a book — over tools designed to keep the conversation going.
Ask the training question: does this app use my kid's words and creations to train models, and did anyone get my consent?
Talk about it — be one of the 6 in 10. Ten minutes of 'AI makes things up sometimes, and it's software, not a friend' outperforms any filter.

Try saying this

“AI is a tool, like a really fancy art supply — it's great at making things WITH you. It's not a person, it doesn't have feelings, and it doesn't get to know your secrets. If any app ever feels like a friend, that's the day you come tell me.”

Questions parents ask next

What age should kids start using AI at all?

Reviewers put the floor for chatbots at no use under 5 and adult-accompanied use from 6–12. Creation-first tools built for children shift the question from age to setup: with a parent-owned account, moderation, and private-by-default work, supervised creative use is reasonable through the early elementary years — which is younger than any chatbot should appear.

How do I recognize a 'companion' app before my kid does?

Look for the tells in the store listing: characters with names and faces, 'always here for you' language, romance or roleplay tags, streaks and daily check-ins. If the product's promise is a relationship rather than an output, it's a companion — and it's not for minors.

Is AI art generation safe for kids?

It's the right category with the wrong defaults in most tools: general art generators are adult products (public galleries, unguaranteed filters, 13+ terms — see our Midjourney review). Kid-safe AI art means moderation on every prompt and image, private work, and a grown-up owning the account. That combination exists; it's just not the default.

What laws actually protect my kid here?

COPPA governs data for under-13s — its 2025 amendments require separate parental consent before a child's data trains AI models. California's SB 243 regulates companion chatbots as of January 2026, and the federal GUARD Act (which would ban AI companions for minors) has advanced through Senate committees but is not law yet. Content safety beyond data remains mostly the parent's job — which is what the category sort above is for.

Sources

Researched August 19, 2026. Products change; we re-check and re-date this article when they do.

The lower-risk shape, built out

Kidgeni is category 3 on purpose: creation-first AI where kids make art, stories, and books — moderated end to end, private by default, with a grown-up owning the account.

See how Kidgeni works