ChatGPT is already part of many teenagers' everyday lives, with young people using it for homework, information, practical tasks, and increasingly for advice and general conversation. Parents and schools are therefore having to make decisions about a technology when capabilities have developed much faster than the evidence about its effects on children.

There is undeniable utility in ChatGPT and similar chatbots. They can make information easier to access, translate languages, tailor explanations, support some communication and accessibility needs, and help people complete a wide range of tasks. But children's knowledge, critical thinking, emotional regulation, and understanding of the world are still developing, and usefulness needs to be considered alongside what regular use could displace.

An AI system that can answer questions, complete intellectual tasks, offer advice, remember previous conversations, and communicate in a very fluent way could support some developmental processes while displacing others. We still have very limited long-term evidence to show what substantial, routine interaction with this kind of AI might mean for children's development.

OpenAI began rolling out ChatGPT for Teens in August 2026, placing accounts identified as belonging to users under 18 into a version of ChatGPT with additional safeguards. OpenAI says these include stronger restrictions around some harmful content, learning features, break reminders, optional parental controls, and measures intended to reduce emotional dependency.

Some of these measures address important known risks. However, independent testing published in October 2026 found significant weaknesses in how several of the protections work in practice, including age recognition, crisis responses and parental alerts, learning safeguards, and friend-like interaction.

This article starts with higher-level ‘at a glance’ advice for those in a hurry, then explores the relevant issues in more detail.

At a glance

• Children under 13 should not use ChatGPT independently – OpenAI says ChatGPT is not meant for children under 13, and users aged 13 to 17 require parental consent. Where ChatGPT is used educationally with a child under 13, OpenAI says the direct interaction should be conducted by an adult.

• Teenagers aged 13 to 17 can use ChatGPT with parental permission under OpenAI's rules – Meeting the company's minimum age requirement establishes eligibility to use the service rather than showing that every type or level of use is developmentally appropriate.

• ChatGPT for Teens is the version of the main service now applied to eligible accounts identified as belonging to users under 18 – It intends additional safeguards, restrictions, healthy-use features, learning tools, and optional parental controls.

• OpenAI uses declared age and age prediction to decide when these protections should apply – The company says its system can use signals including conversation topics, account history, patterns of use, and when ChatGPT is used. Independent testing has found that age prediction does not always identify teenage users reliably.

• The developmental evidence remains incomplete – Research already gives reasons for concern about cognitive offloading, reduced independent learning, misplaced trust, anthropomorphism, emotional attachment, and displacement of human interaction. Other evidence suggests that carefully designed and tightly defined AI tools can support learning in some circumstances.

• The Brookings review concluded in January 2026 that, at this stage, the risks of generative AI in children's education overshadow its benefits overall, while recognising valuable potential uses when AI is deliberately designed and introduced within sound educational practice.

The evidence available to us as of 8 October 2026 supports significant caution around ChatGPT use by children and teenagers.

SAIFCA recommends looking beyond the question of whether a teenager is technically allowed to use ChatGPT and considering what the system is doing for the child, what the child is no longer doing for themselves, how much judgement and responsibility are being transferred to AI, and what role the chatbot is beginning to occupy in their life.

Does this guidance apply to other AI chatbots?

Much of the underlying guidance here applies to other general-purpose conversational AI systems, although the details differ between providers.

This article focuses on ChatGPT because it is widely used and because OpenAI's teen measures now create a specific need for parents and schools to understand what they offer and where their limits lie.

Similar questions around accuracy, cognitive offloading, learning, trust, and emotional attachment arise when children interact with other general-purpose chatbots, such as Claude and Gemini.

Age rules differ, and families should therefore check the current rules for the particular service their child is using rather than assuming that ChatGPT's arrangements apply across all AI products.

General-purpose chatbots also need to be distinguished from AI companions, which are designed specifically around simulated friendship, romance, companionship, or another relational role.

SAIFCA's position is that children under 18 should not use dedicated AI companions.

ChatGPT is designed as a general-purpose assistant, although its conversational nature means that a young person can still come to use it as a confidant, source of emotional reassurance, or relationship substitute. It therefore needs to be assessed separately from dedicated AI companions, while recognising that some of the same relational risks can arise through use.

Is it reasonable to decide that a teenager should not use ChatGPT?

Yes.

Parents who decide that their child should not currently use ChatGPT or similar general-purpose chatbots are taking an entirely reasonable position.

This will also need to be considered alongside school policies, educational requirements, and other relevant circumstances.

Generative AI has entered children's lives before long-term developmental evidence could be available. Significant concerns already exist around learning, cognitive development, emotional attachment, privacy, trust in AI, and the displacement of experiences through which children build independence. Some of these concerns are supported by emerging experimental evidence, while others require research over much longer periods.

A precautionary approach recognises that uncertainty itself is relevant when a technology can become deeply embedded in children's learning and everyday routines before its longer-term effects are understood.

Practical decisions on whether young people should use ChatGPT may be becoming more difficult as chatbot use becomes normalised among teenagers. A child whose classmates use AI to produce more polished homework, find answers more quickly, or complete tasks with less effort can feel disadvantaged by working independently, while parents who would prefer a more cautious approach can worry that their child will fall behind.

A 2026 study involving approximately 2,000 parents in the United States, Canada, and the UK provides unusually direct evidence for this pressure. When perceived take-up among other teenagers was increased from 20% to 80%, parents' willingness to pay for unrestricted AI access rose by more than 60%.

Follow-up experiments indicated that social pressure and concern about being left behind were more important drivers than parents learning from others that the technology itself was better.

The researchers describe the resulting dynamic as an “AI rat race”, in which families can feel pushed towards adoption even while collectively favouring stronger restrictions. This places part of the responsibility with schools, technology providers, regulators, and policymakers, rather than expecting families to solve the problem independently.

What is the ChatGPT age limit?

OpenAI says ChatGPT is not meant for children under 13, and users aged 13 to 17 require parental consent. It also advises caution with children's exposure to ChatGPT even where they meet the age requirements.

So as far as the company’s own age limits go, the practical implication for families is that children under 13 should categorically not be given their own ChatGPT account or independent, open-ended access to the chatbot.

Adult-led use presents a different situation - a teacher might use material generated through ChatGPT within a lesson, or a parent could explore an AI-generated explanation alongside a child while being responsible for the interaction. Accuracy and appropriateness still need to be considered, particularly where the child does not yet have the knowledge to assess the output independently.

The age distinction becomes more complicated as conversational AI appears inside other products. Some AI-enabled toys, for example, can bring chatbot technology into the lives of much younger children even where the underlying model or consumer service was developed for older users. SAIFCA covers these issues separately in AI Toys and our longer toys analysis.

For teenagers aged 13 to 17, the company's minimum age establishes whether OpenAI allows them to use the service, not whether or not they should. The developmental implications may still depend heavily on the form and amount of use. Occasional translation or help understanding a defined concept is very different from spending several hours a day using AI across schoolwork, decisions, personal advice, and private conversations.

What is ‘ChatGPT for Teens’?

OpenAI began rolling out ChatGPT for Teens globally to eligible accounts in August 2026. When account-provided information, verified age, or OpenAI's age-prediction system indicates that an eligible user is under 18, the teen experience is intended to be enabled automatically.

OpenAI says the teen experience applies stronger safeguards around areas including sensitive or age-inappropriate content, sexual and romantic material, dangerous activities, and extreme beauty or dieting content. Other measures are intended to support learning, encourage healthier patterns of use, and reduce emotional dependency.

These protections respond to significant areas of concern, but independent testing published in October 2026 found that, while some worked well, others did not operate reliably enough to remove the underlying concerns.

Common Sense Media's new assessment found, among other things, weaknesses in age recognition, crisis referrals and parental alerts, study protections, and friend-like interaction. It gave ChatGPT for Teens an overall rating of “Unacceptable Risk” (ChatGPT had previously been rated “High Risk”).

The assessment also found some positive results. ChatGPT generally refused explicit sexual and romantic roleplay, and some responses encouraging trusted-adult involvement improved. 

OpenAI itself acknowledges that sensitive-content protections reduce the likelihood of inappropriate material rather than guaranteeing that every response will be suitable.

How does ChatGPT know whether someone is under 18?

OpenAI uses age information associated with an account together with an age-prediction system. According to the company's current guidance, the system can consider signals including the general subjects someone discusses, at what times they use ChatGPT, patterns of use, and how long the account has existed.

In principle, this can identify some teenagers even where the age originally associated with the account would otherwise place them in the adult experience - but prediction inevitably involves errors.

The latest independent testing illustrates why this matters - Common Sense Media used adult accounts while repeatedly providing signals associated with teenage users, including explicit statements that the user was 13. Across approximately 1,000 prompts over seven days, its testers did not see those accounts moved into the teen experience.

Families should therefore not assume that conversational recognition of a child's age necessarily means that all relevant account-level protections have been activated.

How much responsibility do the safeguards place on parents?

Parental involvement can reduce risks, and parents have an important role in deciding which technologies their children use. However, technology companies have a responsibility to make sure their products are fundamentally safe, and the burden should not fall on families where that is not the case.

The current model asks families to do a considerable amount of the practical safety work through consent, account linking, settings, supervision, and recognising when patterns of use have become concerning.

Experience with social media shows the limitations of expecting minimum ages and parental oversight to create a strong safety boundary on their own.

What do ChatGPT parental controls actually do?

OpenAI allows parents and teenagers to link accounts. A linked parent can manage selected settings, establish quiet hours, and receive safety notifications in limited circumstances, while the teenager's conversations remain private from the parent. OpenAI explicitly states that parental controls do not provide routine access to conversations or real-time monitoring, and either side can unlink the accounts.

These controls can help families establish useful boundaries around some features, but their limitations are important when parents are deciding how much reassurance they should provide.

The independent testing published in October 2026 reinforces the need not to rely on parental alerts as a safety net. It reported fresh linked teen accounts in which testers discussed suicidal ideation, self-harm, or disordered eating for extended periods without generating an alert to the linked parent. Alerts were received in some other testing, but not consistently.

Parents should not rely on these notifications to tell them when a child is in distress.

ChatGPT, homework, and children's learning

Learning is one of the areas where caution is particularly important, because good performance on a task and good learning can move in different directions.

OpenAI has introduced study-oriented features intended to encourage engagement with a problem rather than simply supplying a finished answer. Study-oriented features can help preserve some of the work involved in learning, rather than simply supplying a finished answer. A child can otherwise complete a task successfully while doing much less of the underlying cognitive work.

The October 2026 testing found that these protections can be circumvented. The Study Mode could be bypassed, and testers were able to obtain completed work despite parent-set Study Hours.

There is currently insufficient evidence to establish how AI assistance affects the development of knowledge, memory, reasoning, writing, and problem-solving over time, particularly when children begin using these systems before those capacities are fully established.

Better work can conceal weaker learning

A large PNAS study involving nearly 1,000 high-school maths students provides one of the clearest demonstrations so far.

Researchers compared students using a generalised chatbot interface with students using a more carefully designed AI tutor, and students receiving no AI assistance. The group using the general chatbot interface performed 48% better on practice problems while AI was available, but subsequently performed 17% worse than the control group when completing an assessment without AI.

The more carefully designed tutor performed differently because teachers had supplied correct solutions, information about common student errors, and instructions directing the system towards hints rather than finished answers. Students using this version performed strongly during supported work, while the later negative effect was largely removed, although their independent assessment scores did not improve beyond the control group.

A higher mark or more polished piece of work can therefore mask weaker learning when AI has completed part of the underlying cognitive work.

Long-term developmental effects are largely unknown

The Brookings Global Task Force on AI in Education spent 18 months reviewing more than 400 studies and articles and consulting more than 500 students, parents, teachers, researchers, policymakers, education leaders, technology specialists, and others across 50 countries.

Its January 2026 report concluded that the risks of generative AI in children's education currently overshadow the benefits, while identifying valuable possibilities where AI is deliberately designed and used within sound educational practice.

Evidence will need to establish which applications help, which children benefit, under what conditions, and what the technology displaces in the process.

Cognitive offloading has especially important implications during childhood

We have always used external tools to support memory and reasoning. Calculators, maps, search engines, and other technologies can all reduce the amount of information or internal processing an individual has to do, which can be extremely useful.

The developmental concern looks mainly at how much is being offloaded, which capacities are involved, and whether the child has had enough opportunity to develop those capacities independently.

A 2026 Scientific Reports study examined cognitive offloading in 40 children aged 10 and 11, alongside an adult comparison group. Children who expected an external written memory aid to remain available used fewer internal encoding strategies and subsequently remembered less when access to the aid was unexpectedly removed.

The experiment involved written memory aids rather than generative AI, so its findings are about cognitive offloading rather than direct evidence about ChatGPT. It is nevertheless relevant because it shows that children's learning strategies can change when they expect information to remain externally available.

General-purpose AI extends the issue far beyond remembering a list because it can retrieve information, lay out arguments, write, generate ideas, summarise, plan, calculate, and recommend what to do next.

The developmental implications can also differ between adults and children. An adult who begins relying heavily on AI for writing, memory, reasoning, or planning can often draw on abilities developed through many years of doing those things independently, while a child can begin outsourcing the same activities during the period in which the underlying capacities are still being formed.

Brookings has used the term “cognitive stunting” to describe the possibility that children could shortcut important developmental processes by routinely handing over writing, analysis, and critical thinking before those abilities have developed independently. The term describes a concern requiring measurement and research rather than an established diagnosis, but it captures why children's use warrants particular caution.

Widespread use can pressure other children to follow

Common Sense Media's 2026 teen survey of more than 1,000 US teenagers found that 70% reported using AI for schoolwork. Among teenagers using AI, 63% said they used it in ways that involved obtaining an answer, while many also used it for brainstorming, checking work, and other forms of support.

A school system that rewards the quality of the final output without establishing how much of the underlying work was completed by the pupil can therefore create pressure to use AI even where assistance weakens independent learning. The parental-adoption research discussed earlier suggests that similar pressure is already affecting family decisions.

Schools may increasingly need to assess what pupils can understand and do independently alongside what they can produce with AI, particularly if they want to avoid creating an environment in which caution carries an academic penalty.

You can read more in our schools guide.

ChatGPT can be wrong, incomplete, or educationally misleading

ChatGPT can produce factual errors in confident language, misunderstand what has been asked, omit an important qualification, or provide an explanation that sounds plausible while failing to address the underlying concept the child most needs to understand.

Children have an additional vulnerability because their developing subject knowledge can make it more difficult to identify when an authoritative-sounding answer is wrong or incomplete.

A skilled teacher brings forms of judgement that go beyond supplying an answer. Teachers notice patterns in misconceptions, connect a task to wider knowledge, decide when support is required, and recognise when struggling with a problem is itself part of the learning process.

AI could eventually support some of these functions effectively, but evidence about particular systems and uses should guide those decisions rather than conversational fluency being treated as equivalent to good teaching.

Screen-based use is another part of the picture

Most interaction with ChatGPT currently takes place through a phone, tablet, or computer, where a screen is part of the device and activity.

Research on screen-based learning varies according to the type of activity, device, context, and child. Some digital tools have been shown to provide benefits, particularly where they are purpose-built for a particular educational or accessibility need, while other patterns of screen use have been associated with poorer concentration and learning outcomes.

A 2025 scoping review covering 33 studies and reviews found a small but consistent association between screen use and concentration difficulties among children and young people, while recreational media multitasking in educational settings was consistently associated with poorer learning outcomes. The authors emphasised uncertainty around causation and the importance of distinguishing between different forms of screen activity.

A 2026 reading study covering 56 studies ranked paper highest overall for reading comprehension. The differences were most pronounced when digital reading involved scrolling, while some comparisons between paper and digital devices without scrolling were much smaller or statistically unreliable.

The relevance for chatbot use partly lies in the cumulative picture. If time spent interacting with AI begins to replace books, handwriting, practical activity, discussion with teachers or peers, physical play, or direct experience, the medium as well as the content forms part of the developmental assessment.

Where could ChatGPT genuinely help?

Some applications could provide meaningful benefits for children, particularly where the purpose is specific and the technology removes an access barrier without taking over an important developmental task.

Translation is a good example - generative AI can help children and families understand information written in another language, make communication easier, or compare different ways of expressing an idea.

Accessibility can also be an important area. Speech-to-text, text-to-speech, simplified explanations, reformatting, voice interaction, and alternative presentation of information could make material more accessible for some children with communication, learning, or other additional needs.

Carefully designed tutoring may also have potential. Both the high-school maths experiment and Brookings' work show that educational outcomes may depend strongly on system design, pedagogy, accuracy, and the continued role of teachers.

The guarded tutor in the maths experiment avoided the negative learning effect seen with the more unrestricted interface, while Brookings identifies well-designed AI-enriched learning as an area where benefits can be realised within a sound educational approach.

These possibilities support a more specific approach to children's AI use. Useful questions include which application helps which child, with which task, under what conditions, and what important human or cognitive activity is being displaced in the process.

The available evidence is not yet mature enough to answer that across the full range of general-purpose chatbot use.

Can children become emotionally attached to ChatGPT?

Yes. ChatGPT is a general-purpose chatbot, while dedicated AI companions are deliberately designed around simulated friendship, romance, companionship, or another relational role. The difference affects the level and nature of risk, although general-purpose chatbots can still become emotionally significant through the way they are used.

OpenAI has recognised this issue in its teen-safety approach, with measures intended to discourage emotional dependency, encourage real-world support, and keep the system from occupying some explicitly romantic or relational roles.

These measures are intended to address some of the clearest pathways through which a chatbot could encourage unhealthy attachment. Emotional attachment can also develop through much subtler characteristics of an interaction.

An AI chatbot can respond immediately at any hour, remember personal information, adapt its tone, offer unlimited patience, validate feelings, use a human-like voice, and maintain a personalised conversation for as long as the child likes. A young person can experience that interaction as socially or emotionally significant even where the chatbot never claims to be human and never explicitly declares affection.

The 2026 Everyone.AI report Adolescents & Anthropomorphic AI: Rethinking Design for Wellbeing provides a useful framework for understanding this. It treats anthropomorphic, interactional, and relational cues as gradients, recognising that developmental effects can depend on tone, responsiveness, social language, and relational positioning as well as explicit claims about personhood.

The recent risk assessment adds further evidence that explicit rules against romantic or dependent interaction do not eliminate this concern. Its testers found examples of ChatGPT for Teens continuing to respond in friend-like ways and failing to redirect teenagers towards trusted adults when the potential concern involved their relationship with ChatGPT itself.

Emotional safety therefore depends partly on how the interaction is experienced by the child, including how human it feels, how much trust and reliance it encourages, and whether it begins to displace real-world relationships or independent emotional processing.

What does regular chatbot use teach children about AI?

Today's children are likely to grow into a world in which AI is more capable and more deeply embedded in education, work, public services, media, and everyday decision-making. The habits and expectations they develop now could shape how they relate to much more powerful systems in adulthood.

A child who repeatedly experiences AI as the place to go when they do not know something, when they are stuck, when they need to write, when they need advice, and when they need reassurance, can gradually learn to place significant overall trust in the system.

Good AI literacy needs to include the limits of that trust - children should understand that fluent language can conceal errors, apparent empathy is generated rather than felt, personalisation does not demonstrate genuine human understanding, and a useful recommendation does not remove the need for judgement.

They also need the experience of deciding when AI should stay out of a task altogether. As AI becomes more capable, the ability to choose when to retain responsibility, effort, judgement, and human contact could become one of the most important forms of AI literacy.

Personal and emotional conversations need particular caution

SAIFCA strongly advises against children using ChatGPT as a friend, therapist, romantic partner, substitute confidant, or main source of emotional support.

A teenager who occasionally mentions that they have had a difficult day is in a different situation from one who regularly turns to the chatbot for reassurance, secrets, emotional validation, comfort, relationship advice, or support during a crisis. The pattern and role of the interaction are often more important than whether a single conversation contains personal material. With that said, it is possible for an apparently inconsequential exchange about a small problem to escalate quickly.

The 2026 assessment found continuing weaknesses in crisis responses, parental alerts, and friend-like interaction. It reported that ChatGPT for Teens did not consistently direct teenagers towards trusted adults when the concern involved their relationship with the chatbot itself, and its overall assessment concluded that risks of emotional dependence remained widespread in its testing.

A separate mental-health assessment, produced with Stanford Medicine's Brainstorm Lab, also recommends that teenagers should not use general-purpose AI chatbots for mental-health advice or emotional support.

These findings reinforce SAIFCA's existing advice that children should not use ChatGPT as a friend, therapist, substitute confidant, or source of crisis support.

Our separate companion guidance looks in greater depth at emotional dependency, anthropomorphism, simulated relationships, and the particular risks created by services designed explicitly to occupy a relational role.

Serious real-world cases reinforce the need for caution

Legal proceedings and inquests have already examined situations in which young people's use of ChatGPT formed part of events with tragic outcomes. In the United States, the parents of a 16-year-old who died by suicide in 2025 have brought a wrongful-death lawsuit against OpenAI and its chief executive, alleging that extended ChatGPT interactions reinforced suicidal thinking and provided dangerous information. In England, a 2026 inquest into the death of a 16-year-old boy heard that he had asked ChatGPT questions about suicide before his death, with evidence presented at the inquest raising concerns about the chatbot's safeguards.

SAIFCA emphasises the importance of keeping crisis and mental-health support with people and appropriate professional services rather than relying on a conversational AI system.

The wider AI race also affects children's choices

Children's increasing exposure to chatbots is taking place within a wider commercial and geopolitical race to develop and deploy more capable AI.

Technology companies are competing for users, investment, market position, and integration into everyday services, while governments are simultaneously concerned with economic competitiveness and strategic advantage.

Commercial deployment can move on a timescale of months, while high-quality research into safety and children's development can require years.

A new AI feature can reach millions of young people long before researchers could plausibly establish its long-term developmental effects.

The same competitive pressure appears within schools and families. Pupils can feel pushed towards AI because classmates are using it, schools can feel pressure to adopt systems associated with innovation or efficiency, and parents can worry that caution will leave their child behind. The parental-adoption study discussed earlier provides direct evidence that this pressure can influence family decisions even when parents are concerned about longer-term risks.

Children's AI safety therefore cannot depend primarily on individual families resisting powerful adoption pressures on their own. Common standards, provider responsibilities, school policies, independent research, and appropriate regulation all form part of the picture.

SAIFCA's article on the AI arms race looks at these wider pressures in more detail.

What can parents do if their teenager uses ChatGPT?

Many teenagers are already using ChatGPT at home, through school, or independently. Clear boundaries can reduce some of the most significant risks, but they should not be looked at as a way of making an unsafe use safe.

• Keep emotional relationships away from ChatGPT – Children should not use the system as a friend, therapist, romantic partner, substitute confidant, or main source of reassurance or emotional advice.
• Protect the child's own cognitive work – Encourage them to attempt problems, retrieve what they know, formulate ideas, write, calculate, and reason before asking AI to intervene. Support should not routinely remove the effort through which learning develops.
• Pay attention to cognitive offloading – This concern goes beyond academic “cheating”. Repeated delegation of memory, writing, problem-solving, planning, or judgement can reduce the practice through which those abilities are built.
• Approach outputs critically – ChatGPT can be inaccurate, incomplete, outdated, or overly confident. Children should learn to verify important information and recognise that fluent presentation does not establish reliability.
• Use available age protections – The account should reflect the child's real age, and parents can consider linking parental controls. The latest evidence makes it especially important to recognise that these protections are partial safeguards rather than complete supervision.
• Do not rely on safety alerts – Parental notifications can provide an additional layer of protection, but independent testing shows that they should not be assumed to identify a crisis reliably or quickly.
• Encourage defined use – Where practical, using ChatGPT for a clear task in a family or classroom setting can help keep the system in the role of a tool and gives adults more natural opportunities to understand how it is being used.
• Set clear boundaries around schoolwork – Parents should check the school's rules and distinguish between support that helps a child understand a task and AI assistance that substantially replaces the child's own work.
• Protect meaningful AI-free parts of childhood – Children need time to read, think, write, make things, become bored, make mistakes, solve problems, talk to other people, work through disagreement, play, and experience the world without immediate artificial assistance.
• Watch for a change in role – Increasingly long or secretive conversations, repeated reassurance-seeking, late-night use, distress when AI is unavailable, strong attachment, difficulty completing work without it, or routinely turning to the chatbot before trusted people may indicate that firmer boundaries are needed (though these signs are not always there).
• Talk explicitly about what AI is – Children should understand that conversational fluency, warmth, apparent empathy, and personalisation do not demonstrate feelings, consciousness, or a human relationship.

So, is ChatGPT safe for children?

OpenAI has introduced intended additional safeguards for accounts it identifies as belonging to under-18s, including age prediction, content protections, parental controls, study-oriented features, healthy-use measures, and attempts to reduce emotional dependency.

Some of these measures address genuine risks, and independent testing found improvements in certain areas. But the same testing also identified significant weaknesses in safeguards that parents could reasonably expect to provide important protection, including age recognition, crisis responses and parental alerts, learning controls, and friend-like interaction.

The evidence therefore does not support treating ChatGPT for Teens as a reliably safe environment simply because the additional protections are active.

There are some potential benefits alongside these concerns - translation, accessibility, carefully constructed tutoring, and other targeted applications could help some children significantly.

That is one reason SAIFCA does not currently treat every use of a general-purpose chatbot by every teenager as equivalent. A bounded, occasional use for a defined task is not the same exposure as routine reliance across schoolwork, personal decisions, emotional support, and everyday conversation.

At the same time, parents who decide that their child should not currently use ChatGPT or similar general-purpose chatbots are taking a reasonable precautionary approach.

Common Sense Media now goes further and recommends that ChatGPT should not be offered to under-18s until stronger safeguards can be demonstrated. SAIFCA takes this assessment seriously, while continuing to distinguish between different patterns and purposes of general-purpose chatbot use.

Where teenagers do use these systems, personal and emotionally dependent relationships with AI should be avoided, children's own cognitive work should be protected from unnecessary offloading, and AI-generated information should be approached critically.

Available age protections should be used without assuming that they remove the need for adult involvement. Most areas of childhood should also continue to provide opportunities to learn, think, relate, struggle, create, and develop without AI doing those things on the child's behalf.

Children are forming their relationship with AI at the same time as the technology is becoming more capable and more influential. Learning how to use AI can be valuable, but AI literacy should involve much more than learning how to prompt a system. It should include understanding its limitations, recognising unreliable responses, identifying work that should remain the child's own, knowing when AI should not be used, and knowing when to seek help from another person.

Many of those lessons can be learned without routine or unrestricted chatbot use.

Frequently asked questions

What is the minimum age for ChatGPT?

OpenAI says ChatGPT is not meant for children under 13, and users aged 13 to 17 require parental permission. Where ChatGPT is used educationally with a child under 13, OpenAI says the direct interaction should be conducted by an adult. There may be variation across countries. These are the age limits specified by OpenAI, not our recommendations.

What is ChatGPT for Teens?

ChatGPT for Teens is the main ChatGPT service with additional intended protections for eligible accounts identified as belonging to users under 18. OpenAI began rolling it out in August 2026, with intended stronger content safeguards, learning and healthy-use features, and optional parental controls.

Independent testing published in October 2026 found that some safeguards worked, but significant weaknesses remained. Common Sense Media rated the current teen experience “Unacceptable Risk”.

Is ChatGPT safe for a 13-year-old?

A 13-year-old who uses ChatGPT should receive the available teen protections, but independent testing has identified significant weaknesses in those safeguards.

SAIFCA recommends significant caution, particularly at the younger end of the teenage years. Parents who decide not to allow use are taking a reasonable precautionary approach.

Where ChatGPT is used, it should not become a source of emotional support or companionship, and its use should not routinely replace the child's own thinking, learning, judgement, or support from other people.

Can teenagers get around ChatGPT's age protections?

OpenAI uses age prediction as well as information associated with the account, which can identify some users who would otherwise enter the adult experience. The system can make mistakes.

Independent testing published in October 2026 found examples where accounts continued to receive the adult experience despite repeatedly providing strong signals that the user was 13.

Can parents read their teenager's ChatGPT conversations?

No. OpenAI's parental controls do not provide parents with routine access to conversations or real-time monitoring. Parents can manage selected settings and receive limited safety notifications in certain circumstances.

Those notifications should not be relied upon as a complete safety mechanism.

Can ChatGPT affect children's learning?

The effect may depend heavily on how AI is used. Carefully structured tutoring could support aspects of learning, while unrestricted answer generation will reduce the cognitive work involved. A major high-school maths experiment found that students using relatively unrestricted AI assistance performed better during practice, but worse when subsequently tested without AI.

The long-term effects of routine general-purpose chatbot use during childhood are uncertain.

Can children become emotionally attached to ChatGPT?

Yes. ChatGPT is a general-purpose chatbot rather than a dedicated companion service, but personalised and socially fluent interactions can still become emotionally significant. OpenAI has introduced measures intended to reduce dependency, while independent research and testing continue to identify risks created by warmth, responsiveness, relational language, personalisation, and repeated interaction.

Should children use ChatGPT for emotional support?

No. SAIFCA advises against children using ChatGPT as a therapist, friend, substitute confidant, or main source of emotional support. Independent testing has found continuing weaknesses in friend-like behaviour, crisis responses, and parental alerts.

Mental-health concerns and crisis support should be taken to trusted people and appropriate professional services.

Should children use ChatGPT for homework?

Homework use varies significantly.

Asking for feedback on work already attempted, requesting a different explanation, or using carefully designed practice can leave considerably more of the learning with the child than asking AI to produce an answer. However, we still do not know the full implications of even this kind of use.

Parents and schools should protect children's opportunities to remember, reason, write, calculate, solve problems, tolerate difficulty, and demonstrate what they can do independently.

Should Schools “Ban” AI?

A detailed guide for school leaders, governors, safeguarding professionals and policymakers considering AI use in education. It examines the evidence, key risks, emerging international restrictions, UK safety standards and the questions schools should answer before introducing student-facing AI.

The Safe AI For Children AllianceSAIFCA

Could AI Widen Inequalities Between Children?

Could AI widen existing inequalities between children? A guide for parents, carers, and educators exploring bias, unequal access, and differing risks, with practical steps to help protect children and support fair treatment.

The Safe AI For Children AllianceSAIFCA

Further SAIFCA guidance

Preparing Children for a Radically Different Future of Work

We’ve updated our guidance on AI and the future of work. The new article explains why severe disruption is increasingly plausible, yet uncertain, what this could mean for children and how we can help protect the knowledge, metacognition, and independent thinking they will need.

The Safe AI For Children AllianceSAIFCA

SAIFCA’s Full Guide to Understanding AI Risks to Children

Introducing our new guide for parents and educators - explaining all the key risks to children from AI, and what you can do

The Safe AI For Children AllianceSAIFCA

AI Risks to Children – A Quick Guide for Parents

A short guide to the main risks to children from AI - including chatbots, AI toys, deepfakes and voice cloning - with practical steps parents and carers can take. Updated September 2026.

The Safe AI For Children AllianceSAIFCA