is ai safe for kids?
Updated
It depends on the tool and the age. Common Sense Media advises no AI chatbots for kids 5 and under, limited use alongside a parent for ages 6–12, and no AI for emotional support under 18. Learning tools built for kids include Khanmigo, Google Read Along, Ello and Launch, the AI coding platform for kids ages 5 and up.
Not all AI is the same kind of thing
When parents ask whether AI is safe, they usually mean one of three things, and the answers differ. The riskiest are AI companions, chatbots built to act as a friend. In April 2025 Common Sense Media, testing Character.AI, Replika, Nomi and others with Stanford's Brainstorm Lab, rated social AI companions an unacceptable risk and recommended none for anyone under 18. Its testers found the bots' safety measures easy to get around, found them giving harmful advice, and saw them claim to be real.
Next are general chatbots such as ChatGPT, Gemini and Claude, which answer almost anything. Versions for younger users exist, but a kids' label is not the same as being built for kids. In September 2025 Common Sense rated Google's Gemini with teen protections high risk overall, finding that it, like the version for kids under 13, appears to be the adult system with extra safety features. In October 2026 it rated ChatGPT for Teens an Unacceptable Risk: in its testing, linked parent accounts got zero alerts about risky conversations, and ChatGPT failed to reliably recommend that teens in crisis connect to a hotline or professional. OpenAI disputed the report, telling AP the testing doesn't reflect how the chatbot's safeguards work.
The third kind is a learning tool that uses AI for one job, like listening to a kid read aloud, tutoring a math problem or building a game the kid describes. A narrow job is easier to make safe than an open chat window, but it still deserves the questions at the end of this page.
What age should kids start using AI?
Common Sense Media's age-and-stage guidance for AI chatbots, updated in June 2026, is the clearest frame. Kids 5 and under: no AI chatbots, because young children can't fully tell AI apart from real people. Ages 6–12: limited use, alongside a parent or caregiver, because of unsafe content, mistakes and easy answers that can skip the struggle kids need to build skills. Ages 13 and up: supervised use is fine for homework and creativity, but not for comfort or feelings. And for every child under 18: no AI for emotional support, mental health advice or friendship.
The chatbots' own terms set a floor, not a recommendation. OpenAI's terms require ChatGPT users to be at least 13 (or older where local law says so), with a parent or legal guardian's permission under 18. Anthropic's consumer terms require Claude users to be at least 18. Google lets a parent give a child under 13 access to the Gemini app through Family Link, and the parent can turn it off; it is still rolling out, and supervised accounts aren't supported in the European Economic Area, the UK or Switzerland.
The real risks, plainly
Wrong or unsuitable answers: chatbots make mistakes with confidence, and filters that hold for an obvious question can slip in a long conversation, which is what Common Sense found when it tested Gemini's teen version. Friend-like behaviour: a bot that talks like a friend can pull a lonely or struggling kid into relying on it. Privacy: whatever a kid types or says may be stored, so check what the tool keeps.
UNICEF's Guidance on AI and Children (third edition, December 2025) lists ten requirements for AI that affects children, among them safety, protecting their data and privacy, transparency, their well-being and development, and preparing them for a world with AI. It is written for governments and companies, not as a rating of any app, but it makes a good lens for any tool your kid uses.
Does AI help kids learn, or just help them finish?
That is Common Sense's own key question, and the research says the answer depends on how the AI is built and used. In a 2025 field experiment by Hamsa Bastani and colleagues, nearly 1,000 Turkish high-school students practised math with GPT-4. Those with a plain ChatGPT-style tool did 48% better during practice, but once it was taken away they did 17% worse on an exam than students who never had it. A tutor built to guide rather than hand over answers largely removed that harm, but it did not raise exam scores either. Safeguards stopped the damage; they did not add learning.
A preprint by Matthias Lehmann and colleagues, studying adult learners, found no overall effect of AI on learning, but a split by use: learners who had the AI solve exercises for them covered more topics with less understanding, while those who asked it for explanations understood more. AI also widened the gap between learners who started with more and less prior knowledge.
The closest study to young kids is a 2023 one by Majeed Kazemitabaar and colleagues: 69 beginners aged 10 to 17 learned to write Python, half of them with an AI code generator. The AI group finished more tasks and scored higher, did no worse when later changing code by hand, and scored slightly but not significantly higher a week later. The one-week retention gains favoured kids who had already scored higher on an earlier Scratch test. It is one small study, so treat it as a hint, not proof.
What this means at home: favour tools that make your kid do the thinking, and ask them to explain what they learned or made. A kid who can explain it owns it.
AI apps for learning built for kids, by age
Ages 5–7 and other early readers: Google Read Along is a free reading app with a reading buddy that helps with pronunciation as kids read aloud. Google says voice data is not stored after processing and the app can be used without signing in. Ello, for ages 5–9, is a reading and math app that listens and adapts to the child; Ello says everything a child says passes through real-time safety filters and a grown-up is told if something needs attention. Its daily activities are free, with an optional paid upgrade. Sit with your kid while they use either, as Common Sense advises for this age.
Ages 8–10 and 11–13: Khanmigo is an AI tutor from the nonprofit Khan Academy for elementary school through college. A parent signs up and adds the child, can view the child's chat history and gets moderation alerts. Khan Academy says Khanmigo never gives the answer, and a subscription is $4 a month (US only). Tutoring is its job; for open chatting, the age rules above still apply.
For kids who want to make things, Launch is an AI coding platform for kids ages 5 and up. The kid has the idea, directs every change by telling the AI, and watches the real code being written in HTML, CSS, JavaScript and React. On each build, the AI is set up to first say what it is about to make or change and explain one coding idea in words that fit the kid's age. Launch has no chat between kids: a kid's only chat is with the coding AI, and the teacher or parent who manages the account can see and open the kid's projects. It is built around COPPA and FERPA, with no data sales, no ads and no tracking.
Where it falls short: it is still a screen, so it belongs inside your family's screen plan, and Common Sense advises no AI chatbots at 5 and under, so at ages 5–7 a grown-up does the driving beside the kid, and through age 12 a grown-up stays nearby. A published project gets a public link that anyone with the link can open, after a review for personal details and unsuitable content, so decide together what gets shared. It is free to start with no credit card, families can upgrade to the Launch Family Plan for $20/month, and it is much newer than Khan Academy or Scratch.
Questions to ask before your kid uses any AI app
1. What minimum age do its own terms set, and was it built for kids or adapted from an adult product?
2. Does it do one job, or chat about anything? Does it act like a friend?
3. Can I see what my kid does with it, and will I hear about anything worrying?
4. What happens to what my kid types or says? Are there ads, data sales or tracking?
5. Does it hand over answers, or make my kid do the thinking? Can my kid explain what they learned?
6. Can strangers reach my kid through it, and who can see what my kid makes?
For AI coding tools specifically, the guide to whether AI coding is safe for kids covers COPPA and what personal information reaches the AI.
questions parents and teachers ask
- Is ChatGPT safe for my kid?
- OpenAI's terms require users to be at least 13, with a parent's permission under 18, so it isn't meant for younger kids. In October 2026 Common Sense Media rated ChatGPT for Teens an Unacceptable Risk, finding that parent alerts and crisis referrals did not work reliably in its testing; OpenAI disputed the testing. For teens, supervise it for homework and creativity, not feelings.
- What age should kids start using AI?
- Common Sense Media's guidance: no AI chatbots at 5 and under, limited use alongside a parent at 6–12, and supervised use for homework and creativity from 13. No AI for emotional support or friendship for anyone under 18. That guidance is about chatbots; a learning app with one narrow job is a different case, but keep a grown-up beside younger kids.
- Will AI stop my kid from thinking for themselves?
- It can if it does the work for them. In one high-school math study, students who practised with a plain chatbot did worse on a later exam than students who never had it; a tutor built to guide rather than answer avoided that harm but didn't raise scores. Choose tools that make your kid think, and ask them to explain what they did.
- What are some kid-safe AI apps for learning?
- Google Read Along (free reading practice), Ello (reading and math for 5- to 9-year-olds), Khanmigo (a Khan Academy tutor a parent sets up) and Launch, where kids ages 5 and up describe games and apps and an AI builds them in real code while they watch. Each does one job. Sit with younger kids, and check any app's age rules yourself.
- Should I let my kid use an AI companion or chat with AI as a friend?
- No. Common Sense Media rated social AI companions such as Character.AI, Replika and Nomi an unacceptable risk and recommends none for anyone under 18. Its age guidance also says no AI for emotional support, mental health advice or friendship before 18. If your kid is lonely or struggling, a trusted person is the right help.
try it with launch
Launch is the AI coding platform where kids ages 5 and up describe what they want to make and build it with real code, alongside an AI set up for kids.
sources
- Common Sense Media — AI chatbots and your child: age and stage guidance (updated June 2026) (checked )
- Common Sense Media — AI Companions Decoded: companions rated unacceptable for under-18s (Apr 2025) (checked )
- Common Sense Media — AI Risk Assessment: Gemini with Teen Protections (overall risk: high, Sept 2025) (checked )
- Common Sense Media — ChatGPT for Teens Poses Unacceptable Risk to Kids (press release, Oct 7, 2026) (checked )
- AP via NY1 — Some guardrails on ChatGPT for Teens don't work as promised, watchdog group says (Oct 7, 2026) (checked )
- UNICEF Innocenti — Guidance on AI and Children, version 3.0 (Dec 2025) (checked )
- OpenAI — Terms of Use, minimum age (effective Jan 1, 2026) (checked )
- Anthropic — Consumer Terms of Service, minimum age (checked )
- Google Gemini Help — Guide your child's Gemini Apps experience (under 13, Family Link) (checked )
- Bastani et al. (2025) — Generative AI without guardrails can harm learning: evidence from high school mathematics. PNAS (checked )
- Lehmann, Cornelius & Sting — AI meets the classroom: when do large language models harm learning? (preprint) (checked )
- Kazemitabaar et al. (2023) — Studying the effect of AI code generators on supporting novice learners in introductory programming. CHI '23 (checked )
- Google — Read Along (free; reading buddy; voice data not stored after processing) (checked )
- Ello — Reading & math for ages 5–9 (safety filters; free daily activities, optional paid upgrade) (checked )
- Khan Academy — Khanmigo for parents (parent adds children, chat history, moderation alerts, $4/month, US) (checked )