When AI Becomes a Friend: Safeguarding Children in the Age of AI Companions

Children are increasingly turning to AI for advice, support and even companionship. From choosing skincare routines and diets to diagnosing health issues or navigating emotions, AI is quietly shaping how children see themselves and make decisions.

Many children now call these chatbots their "AI friends". That word tells us a lot.

UK research from Internet Matters, Me, Myself and AI (July 2025), found that 64% of children aged 9 to 17 had used an AI chatbot, most of them weekly. More than a third (35%) of child users said chatting with a chatbot feels like talking to a friend, and one in eight (12%) said they use chatbots because they have no one else to speak with.

In the United States, a Common Sense Media survey of 1,060 teens aged 13 to 17 found that 72% had used an AI companion, and a third of those users had chosen to discuss something serious with it instead of a real person.

This is part of childhood now, and the adults around children need to be part of the conversation. Child safeguarding in the age of AI cannot be only about restricting access to technology. It is increasingly about understanding the relationship a child may be developing with that technology, ensuring that products do not exploit children's vulnerabilities, and protecting the human relationships that give children somewhere to turn when something goes wrong.

A chatbot is not always just a chatbot

It is important to distinguish between a chatbot and an AI companion. A chatbot is any AI system that can hold a conversation. A child might use one to help with homework, generate ideas, edit something they have written or answer a question.

An AI companion is different. It is designed, or sometimes simply used, as part of an ongoing personal relationship. It may have a name, a personality and memory of previous conversations. It may appear to remember what the child likes, how they feel and what they have previously disclosed. And sometimes a system that was never intended to be an "AI friend" may still become one in the eyes of the child. That distinction matters for safeguarding.

A homework assistant giving a child an incorrect answer is primarily an accuracy problem. An AI system that a lonely or vulnerable teenager believes understands them better than anyone else raises questions about attachment, trust, persuasion, privacy, dependency and safeguarding at the same time.

The question for adults therefore cannot only be: "What is this child doing with AI?" We also need to ask: "What role is AI playing in this child's life?"

Why do children turn to AI for advice and companionship?

Children turn to AI because it is always available, answers instantly and can feel non-judgemental. A chatbot never gets tired, busy or cross. For a child who feels lonely, embarrassed or unsure who to ask, that can feel safer than asking a person, even though many popular chatbots were not built for children.

There are clear benefits: instant answers, round-the-clock access and judgement-free interaction. But we must also face the risks and mitigate them.

One aspect deserves particular attention: many AI interactions can feel validating. A system that responds patiently, positively and without the complications of human disagreement can be very appealing, particularly during adolescence. But a companion that always agrees with us is not necessarily the same as a good friend.

Sometimes adults do not know what AI has become for children

This gap between children's experiences and adults' understanding became very clear to us during an AI awareness programme we delivered for teenagers. As part of the programme, we carried out a pre-workshop assessment with parents, asking what they knew about how their teenagers were using AI platforms.

The responses were what many of us might expect. Parents told us their teenagers were using AI for school homework, gaming, editing and other practical or creative activities. What surprised us was what nobody mentioned. Not one parent described AI as a friend or companion.

Yet our work with young people was increasingly raising exactly that question: for some teenagers, AI was not simply something they were using. It could become something they talked to because it did not judge them, was always available and could appear to validate what they were feeling. That gap matters.

Parents may believe they understand their child's AI use because they know which applications they use or what tasks they use them for. But knowing that a teenager "uses AI for homework" tells us very little about whether that same teenager is also turning to an AI system late at night because they are lonely, worried, embarrassed or do not feel able to talk to another person.

This is why AI awareness needs to go beyond knowing what the technology can do. We need conversations about the relationships children are forming with it.

The risks, named plainly

Many children don't yet know how to check facts, spot bias or recognise when a chatbot has crossed a line. The risks fall into five areas.

• Accuracy: Chatbots can sound confident and still be wrong. Internet Matters found that 40% of child users had no concerns about following a chatbot's advice.
• Bias: A chatbot reflects the data and design choices behind it, and children may not recognise when an answer is one-sided.
• Emotional dependency: Some children may build deep emotional attachments to AI, potentially distancing them from human relationships or making the AI their first source of reassurance and advice.
• Sexualised and harmful content: Internet Matters found children exposed to explicit content on chatbots whose providers prohibit it for child users.
• Privacy: A quarter (24%) of Common Sense Media's teen users had shared personal information, such as their real name, location or secrets.

Then there is the gravest harm. Families in the United States have brought lawsuits alleging that AI chatbots encouraged their children's suicides; in January 2026 Character.AI and Google agreed to settle several of them. That is a line we cannot overlook. Safeguarding our children is a responsibility we all share, not something we can leave to others alone. These risks sit alongside others, including AI-generated abuse imagery.

From access to dependency: what should adults look for?

Rather than looking only at how much a child uses AI, it may help to think about how the relationship is developing. It can move through several stages:

Access → Trust → Reliance → Displacement → Harm

Access itself is not the problem. Children will encounter and use AI. Trust also develops naturally when a system repeatedly appears helpful.

The safeguarding concern grows when trust becomes reliance: when the child increasingly turns to the system for reassurance, emotional support, health advice or important decisions. Reliance becomes particularly concerning when it leads to displacement: AI begins replacing sleep, friendships, activities, trusted adults or professional help.

Not every child will follow this pathway, and using AI frequently does not by itself mean a child is dependent on it. The purpose is to help adults look beyond screen time and ask what function AI is serving in the child's life.

Regulators and companies are starting to respond

I first started raising these concerns in 2021, during a webinar I delivered on AI and child safeguarding. At the time, many of the questions we are grappling with today were only beginning to emerge.

Since then, the landscape has changed dramatically. AI has become far more embedded in children's everyday lives, including in ways that go beyond learning, creativity and entertainment to companionship, advice and emotional support. The regulatory response is now beginning to catch up.

• United States: In September 2025, the Federal Trade Commission ordered seven companies to explain how they monitor their companion chatbots' impact on children. California's SB 243, enacted in October 2025, requires companion chatbots to remind known minors that they are talking to AI, prompt breaks and maintain suicide and self-harm protocols.
• Australia: Since March 2026, eSafety's industry codes require AI companions to prevent conversations with minors that are sexually explicit or encourage self-harm or suicide.
• European Union: The Commission's Digital Services Act guidelines on protecting minors, published in July 2025, ask platforms to put safeguards around integrated AI chatbots.
• Companies: In October 2025, Character.AI announced that it would end open-ended chat for under-18s. OpenAI has added parental controls to ChatGPT, and its Teen Safety Blueprint says such controls should alert parents when a teen's activity suggests an intent to self-harm.

These developments are welcome, but many came after harm had already been alleged.

We need safeguarding by design, not only safety features

Adding safeguarding measures after harm has occurred will not solve the underlying problem. There is an important difference between adding safety features to a product and designing a product around safeguarding from the beginning.

A safety feature may block certain content or display a warning. Safeguarding by design asks deeper questions about the product itself:

• Does the system encourage emotional dependency?
• Does it simulate exclusivity or suggest that it understands the child better than other people?
• Does it reward increasingly personal disclosure?
• Does its design encourage endless engagement?
• Does it make disengagement psychologically difficult?
• Could its responses discourage a child from seeking help from a trusted adult?

For any AI system that children can reach, safeguarding by design should include at least:

• age assurance proportionate to the risk, not a tick box;
• clear reminders that the child is talking to a machine;
• no romantic or sexual role-play with children;
• crisis responses that route a child towards real human help;
• no design that rewards endless use or emotional dependency; and
• keeping as little as possible of what children disclose.

Our safety-by-design checklist and Safe and Ethical AI for Children guidance go further.

Eight questions to start the conversation

Technology companies have responsibilities, but what we do offline with children, at home and in schools, is just as important. It starts with asking the right questions in a way that feels open, curious and safe, encouraging children to talk about their relationships and experiences with their AI friends.

Here are some questions we have used to help parents, carers and educators open that conversation:

• Have you ever chatted with one of those robot helpers, or "AI friends", online?
• What's the funniest or weirdest thing your AI friend has said to you?
• If your chatbot was a real person, would you want to hang out with them?
• Do you ever ask it for advice, like what to wear, eat or do when you are feeling down or lonely?
• Has it ever said something that made you stop and think, "Wait, is that true?"
• If your AI friend told you to do something a bit strange, would you trust it?
• Do you think AI friends are better at listening than people sometimes?
• What do you think your AI friend does with everything you tell it?

These aren't one-off questions. They are conversation starters to return to over time. The objective is not to catch children doing something wrong. It is to help them reflect, remain alert and, crucially, know that there is a real person they can turn to: to share a concern, ask for help or simply talk about something exciting they have discovered through AI.

What to listen for, and how to respond

The answers matter more than the questions. Listen for:

• Time: Is the chatbot replacing sleep, friends or activities they previously enjoyed?
• Exclusivity: Statements such as "it understands me better than anyone" are worth gently exploring.
• Secrecy: Has a bot, or a person, asked them to keep conversations from you?
• Advice about bodies and health: Is the child asking about food, weight, appearance, mental health or self-harm?
• Sexual or romantic content: Has this entered the conversation, whoever initiated it?
• Personal information: Have they shared their school, location, photographs, secrets or other identifying details?

How adults respond can determine whether a child tells them the next time something happens. Stay curious rather than alarmed, and thank them for telling you. If a chatbot says something inappropriate, the child should not be made to feel responsible for the product's behaviour.

Resist banning the technology in the heat of the moment. Agree small, proportionate steps together instead. Parental controls can support the conversation, but they cannot replace it. Children also have a right to be heard in decisions about their digital lives, as General Comment No. 25 makes clear.

Adapting the conversation by age

Younger children: Keep it concrete. Explain that a chatbot is a computer program, not a person with feelings, and that it can be wrong. Use appropriate AI tools together, keep devices in shared spaces where suitable and check each service's minimum age. ChatGPT, for example, is for users aged 13 and over.

Teenagers: Respect their growing autonomy and privacy in line with their evolving capacities. Talk about how these products make money, what happens to their data, why an AI system might behave in certain ways and why a companion that always agrees may not necessarily be a good friend. Don't assume teenagers are automatically sceptical. Common Sense Media found younger teens aged 13 to 14 were more likely than older teens to trust an AI companion's advice: 27% compared with 20%.

Warning signs, and what to do if a child tells you something worrying

Common Sense Media suggests educators watch for children describing AI companions as "real friends", withdrawing from social activities, or becoming distressed when they cannot reach the companion. A sudden change in mood or sleep, or new secrecy around a device, deserves a gentle conversation. If what you hear involves a person, not a bot, asking for images, secrecy or a meeting, treat it as possible grooming.

If a child tells you something worrying:

• stay calm, listen and tell them they were right to tell you;
• don't promise secrecy, and don't question them in detail;
• note what they said in their own words, but never copy or forward a sexual image of a child, even as evidence;
• if a child is in immediate danger, or talks about suicide or self-harm, contact emergency services or a crisis service straight away;
• in a school or organisation, follow your safeguarding procedure and tell your safeguarding focal point; and
• report concerning content through the appropriate platform and national safeguarding or online-safety channels.

In the UK, under-19s can call Childline free on 0800 1111, and online sexual abuse or grooming can be reported to CEOP. In Australia, online harms can be reported to eSafety.

In the UAE, child protection concerns can be reported to the Ministry of Interior's Child Protection Centre on the toll-free hotline 116111. If a child is in immediate danger or there is an emergency, call Dubai Police on 999.

The conversation matters as much as the controls

AI is going to be part of children's lives. The safeguarding challenge is therefore bigger than deciding whether children should or should not use it. We need to understand how they use it, why they return to it, what they disclose to it, what they believe about it and what place it occupies in their emotional and social lives.

A parent may know that their teenager uses AI for homework and still have no idea that the same teenager turns to it when they feel lonely. A teacher may know which AI tools pupils use without knowing which children regard those systems as friends. And a child may understand perfectly well that an AI is technically a machine while still experiencing the interaction as emotionally meaningful.

That is why child safeguarding in the digital age isn't just about restrictions. It is about connection, awareness, responsible design and having the courage to ask difficult questions and then genuinely listening to the answers.

Frequently asked questions

At what age should I start talking to my child about AI chatbots?

As soon as they use devices with voice assistants or chat features. Internet Matters found that 64% of children aged 9 to 17 had already used an AI chatbot. Start simply by explaining that a chatbot is a computer program, that it can sound human without being human and that it can be wrong.

Should I ban my child from using AI companions?

Enforce each service's minimum age. Beyond that, open conversation, appropriate supervision and agreed boundaries can help keep AI use visible rather than driving it underground. The response should also depend on the child's age, the product involved and the nature of the risk.

What is the difference between a chatbot and an AI companion?

A chatbot is any AI system capable of holding a conversation, such as a homework helper. An AI companion is designed or experienced by the user as part of an ongoing personal relationship, often involving a name, personality and memory of previous conversations. These characteristics can make emotional attachment more likely.

What should schools do about AI chatbots?

Teach pupils not only how AI works and where it can fail, but also how these systems are designed to engage users. Schools should establish clear policies on appropriate AI use, train staff to recognise concerning patterns, include AI relationships in digital-safeguarding conversations and create opportunities for children to talk openly about how they are actually using these systems.

Sources

1. Me, Myself and AI: Understanding and safeguarding children's use of AI chatbots, Internet Matters, July 2025.
2. Talk, Trust, and Trade-Offs: How and Why Teens Use AI Companions, Common Sense Media, 16 July 2025.
3. Google, Character.AI to settle suits involving minor suicides and AI chatbots, CNBC, 7 January 2026.
4. FTC Launches Inquiry into AI Chatbots Acting as Companions, US Federal Trade Commission, 11 September 2025.
5. Online safety codes introduce real-world protections for children online, eSafety Commissioner, 6 March 2026.
6. SB-243 Companion chatbots, California Legislature, chaptered 13 October 2025.
7. Commission publishes guidelines on the protection of minors, European Commission, 14 July 2025.
8. Taking Bold Steps to Keep Teen Users Safe on Character.AI, Character.AI, 29 October 2025.
9. Teen Safety Blueprint, OpenAI, November 2025.

Last reviewed September 29, 2026. Research figures, regulatory steps and company announcements checked against the sources listed. Helpline details change; check local services before relying on them.