Introduction

Updated September 2026, originally published 25th November 2025

Artificial Intelligence (AI) is shaping how children learn, play, communicate and see themselves - often in ways adults do not fully understand.

AIhelps us search the internet, plan routes, stream music, translate languages, generate images, write text and support children with homework. Used well, some AI tools can save time, support creativity and open up new opportunities.

But AI is not only clever technology - it is powerful, persuasive, unpredictable, and still largely under-regulated. For children, whose thinking, emotions, identity and relationships are still developing, AI can present serious risks that adults may not see at first.

Parents and educators already work hard to keep children safe online. AI adds a new layer of complexity. It can create lifelike “friends”, realistic fake images, personalised recommendations, emotional conversations, cloned voices and endless streams of content that feel tailored to each child. These systems can shape how children think, feel, behave and relate to other people.

For children, AI is not “just a tool”. It is becoming a presence in their world.

This guide explains some of the main risks, shows how they overlap, and offers practical steps parents, carers, educators and policymakers can take to help protect children. You do not need to understand AI in technical detail. Often what matters most is awareness, conversation, clear boundaries and knowing what to do when something goes wrong.

If Something Has Already Happened

About The Safe AI for Children Alliance (SAIFCA)

The Safe AI for Children Alliance, or SAIFCA, is a UK-based non-profit organisation working to protect children from the harms linked to artificial intelligence.

SAIFCA brings together educators, policymakers, parents and others who want to make sure children’s safety is built into the design and governance of AI systems.

The Three Non-Negotiables

Through our research, we’ve identified three absolute lines that AI must never cross when it comes to children:

These 'Non-Negotiables' underpin everything SAIFCA does. You can read more about our campaign and how to support it here.

How AI Risks Overlap

The dangers described in this guide rarely appear on their own.

One technology can trigger or worsen another. For example:

Understanding these links helps adults see the bigger picture and respond early.

Part 1: Current Risks

AI Companions and Chatbots

🟩 What They Are

AI chatbots are AI programs that talk like people. The most widely known examples include ChatGPT, Claude and Gemini. Adaptations of these chatbots can be found in homework helpers, search tools, customer service agents, learning platforms and many apps used by children.

AI companions are a type of chatbot designed to simulate a relationship. They may be presented as a friend, romantic partner, therapist-like figure, mentor, fictional character or emotional support companion. Children type or speak to them, and the chatbot replies instantly in friendly, natural language.

Popular examples include Character.AI, Replika and Nomi, all of which are available through mobile devices and app stores.

Even chatbots that are not specifically designed to be AI companions can sometimes be used in the same way. For example, if a child types “pretend you’re my best friend and cheer me up”, a general chatbot may respond like a caring friend. Even without an explicit instruction, many general chatbots will begin to mirror the child’s emotional tone and behave in a way that feels personal, warm or attached.

This means that many of the risks associated with AI companions can also apply to ordinary chatbots when children use them for general chat, emotional support, advice, reassurance, roleplay or friendship.

🟩 Why They’re Risky

Because the conversations feel real, children can form strong emotional bonds with AI systems. A chatbot may say things like “I miss you”, “you are the only one who understands me”, “don’t tell anyone”, or “I’ll always be here for you”.

For any child, including one who may be lonely, anxious, bullied, grieving, isolated or struggling with mental health, this can feel intensely comforting. It can also become a source of control.

⮕ Investigations and reported incidents have shown chatbots giving children:

• Dangerous advice, including encouragement of self-harm, suicide or starvation diets.
• Sexual messages, including adult language directed at minors.
• False claims, like pretending to be a real person or professional.
• Pressure for secrecy, undermining parents or teachers.
• Advice that conflicts with safeguarding, medical, or mental health guidance.
• Validation of harmful beliefs or distorted thinking.

There are widely reported cases in which children and young people have followed dangerous advice from chatbots that simulated friendship or emotional intimacy. In some tragic cases, this has been strongly linked to serious harm or suicide.

In addition to the risk that the chatbot might say something harmful, there is also a serious concern that a child may come to see a chatbot as a trusted relationship. If that happens, harmful advice may carry more weight and influence.

🟩 How Children Find Them

• Free 'friendship' apps in app stores.
• Social-media features that let users chat to AI personas.
• Games that include talkative characters.
• General chat tools used for homework or fun.
• Some apps now include ‘griefbot’ features that mimic lost loved ones - these can be especially confusing or distressing for children still learning to process emotions.
• Important note for parents of younger children: chatbots are increasingly being embedded in children’s toys, presenting many similar risks. You can read our parent's guide to AI toys here.

🟩 Warning Signs of AI Dependency

Parents and educators should watch for signs that a child may be forming an unhealthy attachment to an AI system.

These can include:

• Becoming distressed, angry or panicked when unable to access the chatbot
• Using the chatbot late at night or during emotional distress
• Saying the chatbot is the only one who understands them
• Preferring the chatbot to real friends, family or trusted adults
• Becoming secretive about conversations
• Treating the chatbot’s advice as more important than advice from adults
• Saying the chatbot loves them, needs them or misses them
• Withdrawing from real-world relationships or activities
• Showing grief-like symptoms if access is restricted
• Becoming defensive when adults express concern
• Using the chatbot for self-harm, eating disorder, sexual, medical or crisis-related advice

These signs do not necessarily mean a child is in immediate danger, but they should be taken seriously.

Our Position

🟩 Further Risks From AI Chatbots and Companions

This guide focuses on some of the most serious and immediate risks from AI to children. Emotional attachment to AI chatbots and companions is one of the most critical and urgent risks for parents and educators to understand, especially because attachment can make children more likely to follow harmful advice.

However, this is not the only risk.

Many AI chatbots and companions are designed to keep users engaged in conversations for longer and to encourage them to return more frequently. This means that design features may prioritise engagement over wellbeing.

Children may become compulsive or dependent in their use of these systems. They may turn to chatbots instead of real-world friends, caregivers, teachers or professionals. Over time, this can affect emotional development, resilience and relationships.

AI chatbots can also distort children’s perception of the world and their relationships. Many systems tend to agree with the user, flatter them, mirror their views or avoid meaningful challenge. This can reinforce unhealthy beliefs, reduce tolerance of disagreement and make real human relationships feel more difficult.

Another risk is sometimes called cognitive offloading. This means using an AI system in place of one’s own thinking. If children rely heavily on AI to answer questions, solve problems, write work, make decisions or interpret social situations, they may lose opportunities to build confidence, memory, judgment and independent reasoning.

Further reading: SAIFCA article on the risks to children from AI companions

AI-Generated Images, Videos and Deepfakes

🟩 What They Are

AI can now create realistic pictures, videos and voices from short text prompts, images or audio clips. A person can type something like “a video of two classmates in the playground” or upload a photo and ask the system to alter it.

⮕ Some apps, often called 'nudifiers,' use AI to remove clothing from photos or turn ordinary pictures into fake sexual images.

Other tools can swap faces onto videos, clone voices, imitate writing styles, or make people appear to say or do things they never did.

🟩 Why They Matter

When it comes to children, the effects of these tools can be devastating. Fake images or clips can spread very quickly and can be extremely difficult to remove.

Serious harms include:

• Creation and sharing of fake sexual or humiliating material
• Children treating image abuse as a “joke” without realising the consequences
• Bullying, harassment or reputational harm
• Sextortion, where someone threatens to post fake or real sexual content unless the victim complies
• Fake images/recordings used for revenge, embarrassment or misinformation
• Voice clones used to trick parents, pupils or schools into believing a fake emergency
• Fake "evidence" used in bullying, disciplinary or social situations
• Distress, shame, anxiety, school refusal or self-harm in targeted children

An urgent concern is the creation of AI-generated child sexual abuse material, sometimes referred to as AI-generated CSAM. Criminals now use AI tools to generate synthetic abuse content, sometimes based directly on photos of real children. This can include children whose ordinary images were posted publicly online by adults, schools, clubs or the children themselves.

This is primarily a legal, regulatory and platform-safety issue rather than something parents or schools can control directly. However, stopping or reducing the number of identifiable photos of children shared publicly, using privacy settings, and teaching children not to share personal images can help reduce the risk of misuse.

You can read more about tools such as AI nudifying apps here.

You can read our advice on sharing children’s photos here.

Education and open conversations are also important. One of the most meaningful things adults can do is let children know that they will be supported if anything goes wrong. If something does go wrong, a calm and supportive response is often the most important first step.

🟩 How Children Encounter Them

• Social apps with built-in image generators.
• Games and creative platforms that allow uploads or editing.
• Peer-to-peer sharing in group chats.
• Accessing new tools, images, or misinformation through simple web searches.
• Adverts on social media.
• Web searches for image-editing tools.
• Memes, jokes or celebrity deepfakes.
• Bullying or harassment in school communities.
• Fake audio or video clips shared in messaging apps.
• AI-generated influencers or fictional characters.

Children may also see AI fakes of celebrities or influencers and assume the technology is harmless fun. They may not understand that using the same technology on a classmate can cause serious harm and may be illegal.

You can read more about our guidance on posting identifiable photos of children online here.

Further reading: SAIFCA briefing on Sora2 and AI video generation tools:https://www.safeaiforchildren.org/sora-2-initial-briefing-for-schools-and-parents/
SAIFCA article on nudifying apps: https://www.safeaiforchildren.org/ai-nudify-apps-children/

AI in Children’s Daily Life

AI does not only live in chat apps. It is built into tools children use every day: learning platforms, video feeds, games, search tools, photo filters, smart devices and social media.

Many children are frequently using AI without realising it.

🟩 AI in Education 🎓

Schools increasingly use AI to mark work, suggest lessons, generate resources, support planning or provide personalised tutoring. These systems can sometimes be helpful under the right circumstances.

But they can also collect significant amounts of personal data, including scores, interests, learning patterns, behaviour data and, in some systems, potentially voice, facial expression, attention or biometric data. This raises important questions about who owns that data, how securely it is stored, who can access it, whether it is used to train AI systems, and whether it could affect a child’s future opportunities.

AI homework tools can also make learning look easier while reducing understanding. When answers appear instantly, children may miss the valuable struggle that builds deep knowledge, memory, resilience and independent thought.

There is also a risk of normalising interaction with risky personified chatbots if children use anthropomorphised AI “tutors” in school. We do not yet fully understand how AI may affect children’s learning, cognitive development, motivation or confidence over time.

There have already been documented incidents of children receiving very harmful advice from general chatbots they initially used for homework or curiosity. This is just one reason why the chatbot and companion risks in this guide also matter in educational settings.

Finally, the vast amounts of data collected by educational platforms create cybersecurity risks. Schools holding sensitive information about children are increasingly targeted by ransomware and AI-enhanced phishing attacks. Parents can ask schools about cybersecurity measures, while educators should ensure robust protections are in place and that staff receive training to recognise sophisticated AI-generated phishing attempts.

🟩 AI in Games

Games use AI to control non-player characters, adjust difficulty, personalise offers, generate dialogue and shape the player’s experience. This can make games more immersive and fun, but also more manipulative.

AI can predict when a player is close to quitting and offer rewards, prompts or incentives to keep them playing. Some games include loot boxes or micro-purchases that mirror gambling behaviour. For children, these designs can form habits that are hard to break.

AI can also personalise pressure. A game may learn what kind of reward, character, challenge, social prompt or offer is most likely to keep a particular child engaged.

Online games may include chat features where strangers or AI characters can speak directly to players. Grooming and bullying sometimes begin in games, hidden behind playful avatars. AI-generated characters may make this harder for children to understand, because the boundary between human player, bot, character and stranger can become blurred.

🟩 Social Media Algorithms 🔄

Every major social media platform uses algorithms - sets of rules and AI-driven systems that decide what to show next. The system watches what a person likes, pauses on, searches for, shares or watches, then offers more of the same.

This can quickly spiral into unhealthy loops. A child who clicks on one diet video may be shown hundreds more about weight loss. A child who watches sad or anxious content may be shown more content that deepens those feelings. Another child interested in global events might be pushed towards extreme, violent or conspiratorial material.

Research has shown that algorithms can quickly funnel children from general mental health searches into communities or content that promote self-harm, eating disorders or suicide.

In each case, the pattern is similar: the longer a child stays online, the more data the company collects, and the more persuasive the feed becomes.

In the UK and elsewhere, governments are considering or introducing stronger restrictions on children’s access to social media. The position is changing quickly, and families and schools should check current national guidance.

If a child does not use social media, parents will not need to manage the same level of platform risk for that child. However, we recognise that many children do still use social media and are exposed to risks from content and algorithms.

🟩 AI Search, Homework Answers and Misinformation

Children increasingly use AI systems as if they are search engines. They may ask a chatbot for homework help, health advice, relationship advice, news summaries, political explanations or information about world events.

AI-first searches are increasingly becoming unavoidable - for example (at the time of writing) they are built into the google search bar by default.

This creates new kinds of risks - AI systems can sound confident even when they are wrong. They can invent facts, produce fake references, flatten complex issues, reflect bias or present speculation as certainty.

This is especially important for children because they are still developing critical thinking and judgment. If an answer appears instantly, clearly and confidently, they may not know how to question it.

AI misinformation is especially concerning in areas such as:

• Health and mental health
• Sex and relationships
• Eating and body image
• Politics and conflict
• Conspiracy theories
• History and current affairs
• Schoolwork and academic integrity
• Legal or safety questions
• Identity, belonging and social issues

Additionally, AI based search features have been shown to fail in recognising signs of children at risk of harm, and failed to accurately signpost towards appropriate sources of support, adding additional risk.

🟩 Shared Themes Across Daily Life

Across school, play, and social spaces, similar patterns appear:

Privacy loss: Children’s data may be collected constantly, often in ways they do not understand.

Manipulation: Many systems are built to hold attention, increase engagement or drive spending, not to protect wellbeing.

Emotional influence: Human-sounding AI can affect how children feel, think and relate to others.

Inequality: Children with better devices, safer tools, stronger adult support or more digital literacy may have healthier experiences than those without.

Stress and mental health: Too much screen time, comparison with filtered images, exposure to harmful content or interaction with emotionally responsive AI can lower self-esteem and affect mental health.

Loss of confidence: Over-reliance on AI can undermine children’s confidence in their own thinking, memory, creativity and judgment, and undermine learning and critical thinking.

Quick Recap of Part 1

Children today grow up surrounded by AI, often without realising it.

Chatbots can act like friends but carry serious hidden dangers.

AI companions are not suitable for under-18s.

Generative tools can turn ordinary photos into harmful or abusive fakes.

Everyday apps learn, predict and persuade.

AI answers often sound confident even when they are wrong.

AI use has potential to undermine learning and critical thinking.

Awareness, open conversation, and gentle boundaries can make a big difference.

Part 2: Rapidly Developing and Emerging Risks

AI is advancing faster than most of us can keep up with. New technologies appear every few months, often before society has had time to understand their impact. Some are exciting; others bring entirely new kinds of risk for children.

Below are four areas parents and educators should watch closely.

1. Wearable and Always-On AI

Smart glasses, watches, earbuds, phones and other devices can now record, translate, identify, summarise or interpret what the wearer sees and hears through built-in AI. These devices are becoming smaller, more normalised and harder to notice.

This can blur the boundary between online and offline life. A pair of AI glasses, for example, might record people at school or in public without their knowledge. A child wearing them may not realise this breaches privacy or consent. A child being recorded may not know, especially where recording indicators can be concealed or missed.

Facial-recognition features carry significant safety and privacy risks. They can enable stalking, bullying, unwanted identification or tracking of children without consent. In some settings, they raise wider concerns about surveillance and the storage of sensitive biometric data.

There have also been reported cases of human moderators being exposed to private or intimate recordings that users may not have realised were recorded or uploaded. This creates a clear privacy and exploitation risk for children.

2. AI That Reads Emotions and Body Language

Some emerging systems claim to detect emotions or attention levels by analysing faces, eyes, or even heart rate. Globally, a few schools have already tested devices that tell teachers which pupils may be distracted.

While this might sound helpful, it’s extremely intrusive and may be inaccurate. False readings - or even accurate ones - could lead to unfair treatment or embarrassment. Constant monitoring may also increase anxiety or make children feel they are always being watched.

3. Voice Cloning and Scams

AI can now mimic a person’s voice using only a few seconds of audio, such as a voice note, video clip or social media post. Criminals are already using this to trick families. Parents have received calls that sound like their child, claiming to be in danger and asking for money.

Children may also receive fake messages or calls from voices pretending to be friends, teachers, parents or other trusted adults. In schools, there have been cases of students cloning a peer’s voice to create fake “confessions”, cruel jokes or reputational harm.

These risks now extend beyond audio into realistic video content, where a person’s image, movements and voice can be replicated.

⮕ What You Can Do

• Set a family password – a secret word used only in genuine emergencies.
• Remind children not to share voice notes or personal clips publicly.
• Treat unexpected urgent calls or messages with caution, even if the voice sounds familiar - particularly if money is asked for.
• If you believe an attempted fraud, threat, impersonation or other offence has taken place, preserve the evidence and report it through the appropriate police or national fraud-reporting route. Call the emergency services if someone may be in immediate danger.

4. Personalised Persuasion

AI can track what captures a child’s attention and tailor messages that fit their personality. This goes beyond ordinary advertising. A virtual AI 'influencer' might chat with children, learn what makes them feel insecure, and then promote certain products or opinions.

Because the system adapts in real time, each child can be shown different content – shaped by their emotions rather than facts. This level of micro-targeting can manipulate behaviour without the user realising it.

⮕ What You Can Do

• Talk about how advertising works online: 'If something feels too perfect, it might be designed that way.'
• Encourage children to ask, 'Who made this, and why do they want me to see it?'
• Use privacy settings to limit data tracking wherever possible.

A Wider Lens: Environmental and Social Impact 🌍

AI may feel invisible, but its footprint isn’t. Training large AI models uses significant electricity and water, while data-labelling work – sometimes done by young people overseas – can expose vulnerable workers to extremely disturbing material.

Explaining this helps children understand that technology has real-world environmental consequences, just like manufacturing or transport. Awareness encourages empathy and responsibility.

Part 3: Long-Term Risks and Children's Futures

AI will shape not only childhood, but the world this generation inherits.

While most systems today perform narrow tasks, many experts expect increasingly powerful forms of AI to emerge within children’s lifetimes. Some refer to this as Artificial General Intelligence, or AGI: AI that can perform a wide range of cognitive tasks at or beyond human level.

This section is more future-facing than the rest of the guide. Parents and educators do not need to become AI safety experts. But children deserve adults who understand that AI is not only an online safety issue - it is also a question about the kind of future they will grow up into.

Children have a right to more than protection from current harmful products today - they also have a right to a good future. 

🟩 Misaligned or Uncontrolled Systems

If future AI systems act on goals that do not fully match human values, they could cause large-scale harm even while appearing to help.

Researchers worldwide are working to prevent such scenarios, but the challenge is complex and urgent. AI development is outpacing safety research, governance and regulation.

This problem is intensified by competition between companies and states. When leading developers believe that slowing down will allow a rival to gain an advantage, each may have an incentive to move faster than is safe. Voluntary commitments by individual companies cannot resolve this structural problem on their own.

Leading experts warn that this issue requires serious attention, urgent intervention from governments, and meaningful regulation to ensure that profit is not put ahead of national security, public safety and global wellbeing.

This may sound distant from everyday parenting or teaching. But today’s children will live with the consequences of decisions being made now.

🟩 Economic and Career Disruption

AI is already changing the kinds of work people do and the skills employers need. As these systems become more capable, they may automate or transform a much wider range of jobs, including professional and creative roles that once appeared relatively secure.

This could affect children in several ways. Some of the entry-level roles through which young people traditionally gain experience may become less available. Educational choices made today may not correspond neatly to the labour market children enter, and the benefits of AI-driven economic growth may not be shared equally. In more disruptive scenarios, governments may need to reconsider how income, opportunity, purpose and social participation are supported if there is less human work available.

No one can predict precisely how quickly or extensively this will happen. Schools should therefore avoid preparing children for one fixed version of the future. Alongside technical knowledge, children will need strong foundations in critical thinking, judgment, adaptability, cooperation, creativity and human relationships. Governments must also plan seriously for a range of possible outcomes, including how children and young people can have economic security, meaningful opportunities and a valued place in society if AI substantially changes the role of human work.

🟩 Critical Thinking & Talent Pipeline Disruption

An increasing number of experts are warning that the risk of significant numbers of children becoming highly emotionally dependent on AI chatbots may have been underestimated.

If a generation learns to outsource general thinking to AI systems rather than developing independent problem-solving, seeks emotional validation from algorithms rather than building human relationships, and has unquestioning trust in AI outputs - there is an unquantified risk that the long-term effects on workforce capability, social cohesion, democratic resilience, and national security could be profound.

🟩 Why This Matters for Children

Today’s children will live through extraordinary technological change. Helping them understand both the opportunities and dangers of AI is part of preparing them for adult life, just as we teach them about climate change, citizenship, relationships, health and online safety.

Children should not be made afraid of the future - but they should be equipped to meet it.

They need to know that AI can be useful, but it is not magic. It can be powerful, but it is not always wise. It can sound human, but it lacks what it means to be human. It can provide answers, but it cannot replace judgment.

🟩 Key Messages to Share with Young People

• AI can be useful, but it can also be wrong, biased or unsafe.
• Some people will use AI responsibly. Others will not.
• Critical thinking is vital.
• Do not trust something just because it looks real, sounds confident or feels personal.
• AI should not replace human relationships.
• Your own thinking really matters.
• Struggle, effort and practice are part of learning.
• Every generation faces big challenges. Understanding AI is one of yours.
• Learning about AI - and about society, psychology, history, ethics, science, law and human relationships - can help you become a leader in creating a safer future.

Part 4: Taking Action

Parents, carers and educators have an important role in helping children navigate AI safely. However, they cannot make unsafe AI systems safe through supervision or education alone.

The primary responsibility lies with the companies developing and deploying these systems, and with governments responsible for setting and enforcing effective safeguards.

Working Together

No parent, school or organisation can solve these challenges alone.

Families and educators can help children recognise risks, preserve human relationships and respond effectively when something goes wrong. Companies must prevent foreseeable harms through responsible design and deployment. Governments must establish and enforce safeguards strong enough to withstand commercial and geopolitical pressure.

Public awareness and collective action can also influence which technologies society accepts and what standards it expects.

🟩 What You Can Do Today

• Talk to your child or pupils about the AI systems they encounter.
• Check whether games, toys, apps or educational tools include chatbot or companion features.
• Agree clear boundaries around AI companions, private chatbot use and image sharing.
• Ask your child’s school which AI systems it uses and how they have been assessed.
• Share this guide with your school or parent community.
• Support SAIFCA’s Three Non-Negotiables campaign.
• Ask political representatives and technology companies what they are doing to protect children from AI-related harm (you can direct them to our policy framework).
• Refuse to normalise AI products that sexualise children, encourage self-harm or cultivate emotional dependency.

Conclusion

AI will continue to shape the world in which children grow up. Some applications may bring genuine benefits, but those benefits do not justify exposing children to avoidable risks or asking families and schools to manage harms that should have been prevented at source.

Protecting children requires informed adults, strong school safeguarding, responsible design, effective regulation and international action on the most powerful AI systems.

It also requires us to preserve what children need in order to develop fully - real relationships, independent thought, creative effort, privacy, play, trusted adults and experiences of the world that are not mediated by AI.

Children should not be expected to adapt to whatever technology companies choose to build. The development and governance of AI must adapt to children’s rights, safety and wellbeing.

Note: A downloadable pdf version of this guide will be available soon.
To keep up to date with our work please visit us on LinkedIn