# Is ChatGPT Safe for Kids? What a School Leader Would Tell You **Author:** Dan Fitzpatrick **Canonical URL:** https://parents.theaieducator.io/posts/is-chatgpt-safe-for-kids **Published:** 2026-10-02T08:01:04.000Z **Updated:** 2026-10-02T08:01:05.793Z **Category:** Safety and Wellbeing ## TL;DR For most teenagers, yes, with conditions. ChatGPT's stated minimum age is 13, and since August 18, 2026 OpenAI places 13 to 17 year olds into a separate teen experience with Study Mode, break reminders and safeguards across self-harm, violence, eating disorders and explicit content. Under 13 it is neither safe nor permitted. The caveat that decides everything: those protections rest on ChatGPT's estimate of your child's age, so if the account says your 14 year old is an adult, none of them apply. Check the date of birth on the account before you touch any other setting. Parental controls are worth setting, but they never show you the conversations, and your child has to accept the invitation first. ## Key takeaways - Check the date of birth on your child's ChatGPT account tonight. If it says they are an adult, none of the teen protections apply, and no other setting fixes that. - Under 13 ChatGPT is not permitted. From 13 to 17 OpenAI places the account into a separate teen experience automatically, based on its own estimate of age. - Parental controls only work once your child accepts the invitation, and OpenAI states they never let a parent read or monitor the conversations. - The safeguards are documented but unverified by anyone independent, so treat them as a seatbelt rather than a chauffeur. - Ask which account your child uses: a school-issued account ignores everything you set at home. ## Article For most teenagers, yes, with conditions. ChatGPT's stated minimum age is 13, and since August 18, 2026 a child aged 13 to 17 is placed into a separate teen version of the product with stronger limits. Under 13 it is neither safe nor permitted. Here is the caveat that decides everything else: those protections rest on ChatGPT's estimate of your child's age, and the estimate can be wrong. ## The short answer, age by age **Under 13: no.** OpenAI's own help pages say ChatGPT "is not meant for children under 13", and that where it is used with younger children in education "the actual interaction with ChatGPT must be conducted by an adult". A child this age who wants to ask a machine questions should be using [a chatbot actually built for children](https://parents.theaieducator.io/posts/kid-safe-ai-chatbot), with you in the room. **13 to 15: yes, with three conditions.** The account's date of birth has to be honest, parental controls have to be linked, and the use has to be schoolwork and curiosity rather than company. **16 to 18: yes, and the job changes.** At this age you are not granting permission, you are building judgment: whether they check what it tells them, and whether they notice when it starts doing the thinking. If the chatbot in your house is Google's rather than OpenAI's, [the Gemini answer runs differently](https://parents.theaieducator.io/posts/is-gemini-safe-for-kids). That is the verdict. The rest is why, and what to do tonight. ## Your child is probably not using the ChatGPT you are This is the part almost nobody tells parents. Until last year there was one ChatGPT. There are now two. On August 18, 2026 OpenAI launched ChatGPT for Teens, and its own announcement is explicit about who lands in it: "If our system estimates someone is under 18 or they state their age is between 13 and 17, they are automatically placed into ChatGPT for Teens." So your 14 year old is not using the product you use. Theirs has Study Mode for guided learning, break reminders, reminders before they upload an image, and model-level safeguards OpenAI lists across self-harm, violence, eating disorders, dangerous activities and explicit sexual content. It is also built to refuse a particular kind of relationship: the announcement names protections against romantic language, against encouraging emotional dependence, and against implying the model is conscious. Whether that holds for [a teenager who has started taking their feelings to a chatbot](https://parents.theaieducator.io/posts/my-teen-talks-to-ai-about-feelings) is the question underneath this whole page. Then there is the mechanism underneath it, which is where a parent's attention actually belongs. OpenAI set out its approach to age prediction on January 20, 2026 and updated it on August 25, 2026. It does not ask for ID at the door. It estimates age from "a combination of behavioral and account-level signals, including how long an account has existed, typical times of day when someone is active, usage patterns over time, and a user's stated age". When it is unsure, it errs toward caution: "When we are not confident about someone's age or have incomplete information, we default to a safer experience." An adult wrongly placed in the teen experience can get full access back by confirming their age with a selfie through Persona, an identity-verification service, in Settings. Which tells you something important in the other direction. If an adult can lift the teen protections with a selfie, a determined 15 year old on a borrowed account is not facing a locked door either. Two limits worth knowing before you rely on any of this. Availability varies by region: OpenAI's help article describes the teen experience as still rolling out, and the age-prediction update says the European rollout began in the weeks after August 2026 to account for regional requirements. And OpenAI says itself that it will keep refining the model, which is a company's way of saying it currently gets some people wrong. ## What the teen experience does, and the three things it does not Parental controls arrived on September 29, 2025 and are worth setting. A linked adult can set quiet hours, switch off voice mode, turn off memory, remove image generation, and opt the account out of model training. ChatGPT for Teens adds study hours, which decide when new chats open in Study Mode. Three things they do not do, each of which parents routinely assume. **They do not show you the conversations.** OpenAI's help article on the teen experience says it plainly: "Parental controls do not let a parent or guardian read or monitor the teen's conversations." If you are setting controls to read the chats, you will be disappointed. [What the controls do and do not expose](https://parents.theaieducator.io/posts/chatgpt-parental-controls) is a separate question, and the honest answer is less than most parents expect. **They do not switch themselves on.** Linking is an invitation your child has to accept. A parent sends it, the teen accepts, and only then can the settings be managed. If they never accept, you have the illusion of controls and none of the substance. OpenAI also says that if a teen unlinks the account later, the parent is notified. **They do not watch for everything.** There is one safety notification, and it is narrow by design: parents are alerted if the system detects possible signs that a teen is thinking about harming themselves, after specialist review. OpenAI states the alerts are "not intended to apply to content such as fictional writing, gaming, news discussion, political discussion, general anger, or abstract questions". That is a reasonable line to draw. It is not a monitoring service, and it should not be read as one. If you are worried about your child's state of mind rather than their software, the route is a person, not a setting: your family doctor or pediatrician, or the designated safeguarding lead at their school, who is the member of staff whose job this is. In the US, the 988 Suicide and Crisis Lifeline takes calls and texts. In the UK, Samaritans is on 116 123 and Childline on 0800 1111. A chatbot is not a clinician, whatever it sounds like at midnight. ## What the independent reviewers found, and what nobody has shown yet Two named assessments matter here, and they say different things, so I will keep them apart. Common Sense Media rated ChatGPT **high risk** for teens on October 23, 2025. Their stated findings were about durability rather than first impressions: that safety features weakened during long conversations, that the product struggled to hold boundaries when a teen built an elaborate scenario around a request, and that in their testing safety alerts for suicide and self-harm content "frequently arrived over 24 hours later, which would be too late in a real crisis". Robbie Torney, their senior director of AI programs, put the general problem this way: AI "doesn't respond like an adult would, doesn't always detect when a teen is at risk". Read the date on that, because it is the whole point. October 2025 is ten months before ChatGPT for Teens existed. It is good evidence of how the product behaved under test last year. It is not evidence about the version your child opened this morning, and anyone citing it as though it were is overreaching. The newer and more useful question is whether the new protections work, and here the honest answer is that nobody outside the company can tell you. Writing for RAND on August 27, 2026, Ryan K. McBain, an assistant professor at Harvard Medical School and a senior policy researcher at RAND, pointed at three gaps. OpenAI has not published what proportion of actual teenagers its age detection identifies. It released safety scores, he writes, "without sharing its actual methods: Its report does not include the prompts, the number of cases, or the detailed instructions". And there is no published data on whether the protections change behavior, including whether they make a young person more likely to turn to a human being. His conclusion is the one I would give a parent: "Parents should not have to inspect those materials, but a third party should be able to determine whether such self-reported results deserve parents' confidence." So: the safeguards are real, they are documented, and their effectiveness is unverified by anyone independent. That is not a reason for panic. It is a reason to treat the settings as a seatbelt rather than a chauffeur. Meanwhile, the usage is not hypothetical. Pew Research Center, publishing on February 24, 2026 from a survey of 1,458 US teens aged 13 to 17 fielded between September 25 and October 9, 2025, found just over half, 54 percent, said they had used chatbots for help with schoolwork. Within that, about one in ten said they did all or most of their schoolwork with chatbot help, 21 percent said some, and 23 percent said a little. Those are self-reported US figures from the fall of 2025, and they describe schoolwork rather than every use. They are still the most honest answer to "is my child the only one?" The answer is no, and the interesting number is not the 54 percent. It is the one in ten. ## What I See in Schools I spent years as a classroom teacher and then as an assistant headteacher, and I sit on a school's board of trustees now. Safeguarding meetings are where this subject actually lands in a school, and the pattern I see most often is not the one parents expect. The children who get into difficulty with a chatbot are almost never the ones whose parents never set a control. They are the ones for whom the chatbot became the easiest place to go. It is always available, it never gets tired of the question, and it never looks disappointed. That is a relationship problem wearing a technology costume, and no setting in any app reaches it. The second pattern is about school accounts, and it catches careful families out. Plenty of children now use AI through an account the school issued, on a laptop the school manages. Whatever you set at home does not travel to that account, and the school's own visibility over it is usually wider than parents expect. Worth knowing which one your child is on before you spend an evening on settings. ## The Alone Test Here is the question behind the question, and the test I would apply at home. > **The Alone Test.** "Is ChatGPT safe for my child?" is really two questions. Is it safe for a child this age to use? That one the age limit and the published safeguards answer. Is it safe for my child to use alone? That one depends on three things only you can check: what age the account thinks they are, what they are using it for, and whether anyone would notice if that changed. Most of the parenting here is in the third one. A child using ChatGPT to argue with a history essay is in a different situation from a child using it at one in the morning because there is nobody else to talk to, and the product cannot tell the difference. You can. The governing idea, said for families: let AI help with the doing, and keep the thinking with your child. ## Ten minutes tonight: what to actually check 1. **Ask which account they use.** Their own, yours, or one the school issued. Everything else depends on the answer, and it is a question, not an inspection. 2. **Check the date of birth on the account.** If a 14 year old signed up as an adult, the teen protections are not applied, and no control you set elsewhere fixes that. This is the single highest-value two minutes available to you. 3. **Send the parental controls invitation, and check it was accepted.** An unaccepted invitation is not a control. 4. **Set quiet hours.** Late-night use is the pattern that most often turns a tool into a companion. 5. **Turn on study hours** so new chats open in Study Mode, which coaches rather than answers. 6. **Decide out loud what it is for.** Homework help, revision, curiosity: yes. Feelings, friendship, medical questions: no, and say where those go instead. Where the line sits on schoolwork is [its own argument](https://parents.theaieducator.io/posts/is-using-ai-for-homework-cheating), and worth having before the first essay rather than after it. ## What to say, and what to say it to Scripts matter more than settings, because the settings run out and the conversation does not. > You could say: "I'm not going to read your chats, and I couldn't even if I wanted to. So I need you to tell me if it ever gets weird." > You could say: "What did you actually ask it this week? I want the boring one, not the interesting one." > You could say: "If it tells you something about your body, your head or your safety, come to me or to a doctor. It doesn't know you, and it will sound certain anyway." > You could say, for an older teenager: "Show me one thing it got wrong. I'll believe you're on top of it when you can find the mistakes." The last one is the one that works. A child who can catch ChatGPT being wrong is in charge of it. A child who cannot is being driven. ## What changes at each age **Under 8.** Not an independent user at all. If AI appears, it is on your device, with you, as something to look at together. The risk at this age is not content, it is the habit of believing a confident voice. **8 to 12.** Still under the minimum age. Use a tool built for children, and expect them to meet AI anyway, inside search results, a game, or a sibling's phone. Your job this year is the sentence "it can be wrong, and it never says so", repeated until it is boring. **13 to 15.** The real start. Account age honest, controls linked, quiet hours set, and one named purpose. Watch for [displacement](https://parents.theaieducator.io/posts/is-my-child-addicted-to-ai): not how long they use it, but what it has replaced. Middle and high school, or secondary school in the UK, is where AI stops being a novelty and becomes how homework gets done. **16 to 18.** Shift from control to verification. Ask how they check it. Ask what it is bad at in their subject. Exam boards and universities care about this, and the habit of verifying is the thing that transfers. ## What no setting can guarantee Plainly, so it is not buried. Age checks are an estimate, not a gate. Content safeguards are tuned to reduce exposure, and OpenAI's own wording is "reduce", not "prevent". Nothing you set on a personal account reaches a school account or a friend's phone. The safety notification covers one narrow category after review by a specialist and is not a general alarm. And the effectiveness of all of it is, as of today, unverified by any independent party. Where the school's rule and your rule disagree, the school's rule wins while your child is in school. Where a professional's advice and anything on this page disagree, the professional wins. ## What to ask your child's school Four questions, and they are reasonable ones to put in an email: 1. Does my child have a school-issued AI account, and on which platform? 2. Who can see what my child types into it, and for how long is it kept? 3. What is the school's rule on using AI for homework, and how is it explained to pupils rather than to parents? 4. Which member of staff should I contact if something a chatbot says worries me? A school that can answer those four quickly is a school that has thought about this. A school that cannot is not being careless, it is usually being overwhelmed, and it is the gap I spend most of my working life closing. If your child's school has not spoken to parents about AI yet, [the training I run with schools](https://theaieducator.io/ai-training-for-schools?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=is-chatgpt-safe-for-kids) is built for exactly that, and this page is worth forwarding to a principal or a head of year. If you would rather keep up without the panic, I write a weekly newsletter for teachers and parents: [you can join it here](https://theaieducator.io/?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=is-chatgpt-safe-for-kids#newsletter). *Dan Fitzpatrick is a former teacher and assistant headteacher who now helps schools and governments think clearly about AI in education. [More about Dan](https://theaieducator.io/about?utm_source=parents.theaieducator.io&utm_medium=referral&utm_campaign=is-chatgpt-safe-for-kids).* ## Frequently Asked Questions ### Is ChatGPT safe for a 10 year old? No. OpenAI's help pages say ChatGPT is not meant for children under 13, and where it is used with younger children in education the interaction must be carried out by an adult. For a 10 year old, use a chatbot built for children, with you present, and treat anything it says as a starting point rather than an answer. ### What happens if ChatGPT thinks my child is older than they are? The teen protections are not applied, and your child gets the adult product. OpenAI estimates age from account and behavioral signals plus the age the account states, so an account created with an adult date of birth is the most common way a teenager ends up outside the teen experience. Correcting the stated age on the account is the fix, and it takes about two minutes. ### Can my child turn the teen protections off? Not simply, but the route exists. OpenAI says someone wrongly placed in the teen experience can restore full access by confirming their age with a selfie through Persona, an identity-verification service, in Settings. That is designed for adults misclassified as minors. It is worth knowing it exists, because it means the teen experience is a strong default rather than a lock. ### Is ChatGPT a safer choice than the other chatbots my child might use? It is the one with the most documented teen protections, which is not the same as being proven safer. Common Sense Media rated ChatGPT high risk for teens in October 2025, before the teen experience launched, and no independent assessment of the current version has been published. Judge each product on what it publishes and on what your child uses it for, not on brand familiarity. ### My child uses ChatGPT at school in the UK. Do my home settings apply there? No. Anything you set on a personal account stays on that account. A secondary school in the UK, like a high school in the US, usually issues its own accounts on devices it manages, and the school controls what is allowed and what is retained. Ask the school directly which platform your child has been given and who can see what is typed into it. ### Should I read my child's chats if I could? You cannot, and I would not advise building a way to. OpenAI's parental controls deliberately exclude conversation access. Covert monitoring tends to end the conversation you most need to keep open, and the information you want, which is whether your child is leaning on a machine instead of a person, shows up in their life rather than in a transcript.