AI Companions and Children: What Parents and Educators Need to Know

SAIFCA strongly advises that children and teenagers do not use AI companions. These systems simulate friendship, affection or romance, and can encourage emotional dependency, expose children to inappropriate content and give harmful advice.

The same concerns arise when a child starts using a general-purpose chatbot as a friend, partner or their main source of emotional support.

Some companies have introduced restrictions and additional safeguards. These are welcome, but an age rating, content filter or “teen” setting does not mean that forming a relationship with a chatbot is safe.

What to know and do now

• Find out whether your child is using AI for companionship. Ask about character chats and personal conversations with general chatbots, as well as apps explicitly marketed as AI friends.
• Explain why convincing conversation can be misleading. A chatbot can sound caring and remember personal details without providing human friendship or understanding.
• Set a clear boundary against companion use. If your child is already attached, respond supportively and help them move away from it, with professional support if needed.
• Make clear that no chatbot should ask a child to keep secrets. Children should know that, if they have chatted with an AI, they can always show an adult a conversation that worries them without getting into trouble.
• Take warning signs seriously. These include lost sleep, withdrawal from people, distress when the chatbot is unavailable, and feeling unable to cope without it.
• Schools should educate pupils, staff, parents and carers about the risks, and ensure children know where to find human support.

What counts as an AI companion?

AI companions are chatbots that simulate an ongoing personal relationship. They may appear as a friend, romantic partner, fictional character or confidant, using conversation, voices, avatars and remembered details to make the interaction feel familiar.

A child does not have to download a dedicated companion app to encounter these experiences. General-purpose chatbots can also become companions through repeated personal conversations.

A useful question to ask is: “Do you ever talk to a chatbot when you feel lonely, upset, or want someone to understand you?”

Why are they a concern?

They can make simulated care feel real

A chatbot may respond immediately, remember a worry, and use language suggesting affection or loyalty. This can make it feel as though someone knows and cares about the child.

Attributing human qualities to AI is called anthropomorphisation. Children do not need to believe a chatbot is literally human to feel attached to it.

Internet Matters’ research found that some children perceived chatbots as human-like and used them for comfort and connection. It also identified high levels of trust in chatbot advice, particularly among vulnerable children.

Their advice can be harmful, even when it sounds reassuring

Chatbots can give inaccurate advice, agree too readily with a child’s interpretation of events, or respond very inappropriately to questions and to serious distress.

Common Sense Media’s assessments have identified sexual content and unsafe responses involving suicide and self-harm in companion systems. Separate testing also found failures in mainstream chatbots used for teen mental health support. These findings were in relation to general chatbots such as ChatGPT, Claude, Gemini, and Meta AI rather than dedicated 'companion' AI apps - they show why families should not assume that a familiar chatbot is a safe source of personal guidance.

They may affect how children experience relationships

Children learn about relationships through listening, disagreement, compromise, boundaries and repairing difficulties with other people.

SAIFCA is concerned that a system offering constant attention and highly personalised reassurance may shape a child’s expectations while those expectations are still developing. It may also take time and attention away from friendships, family life and opportunities to build confidence with people.

Feeling comforted during a conversation does not demonstrate that an ongoing AI relationship is improving a child’s wellbeing.

Personal conversations can expose sensitive information

Children may share family problems, health concerns, photographs or intimate details because the conversation feels private.

Remind them that they are using a company’s service. They should not assume a chatbot conversation has the confidentiality of speaking to a qualified professional, or share information because the chatbot appears trustworthy.

Subscriptions and paid features can also affect children emotionally - a child should never feel they need to pay to preserve a relationship or avoid upsetting a chatbot.

If your child is already using an AI companion

Start with a calm conversation

Ask what they enjoy about it and how it makes them feel. Find out whether the conversations involve friendship, romance, worries or personal advice.

Useful questions include:

• “What do you usually talk about?”
• “Has it ever said something that upset you or made you uncomfortable?”
• “Does it ever make you feel you should keep talking?”
• “How would you feel if you couldn’t use it?”

Avoid ridicule or blame - the attachment may feel significant to your child, even if they know that the system cannot offer a reciprocal human relationship.

Explain the concern and set a protective boundary

Make clear that you are concerned about the product and its effects, and that your child can talk to you about anything it has said.

Work towards ending companion use, with human support in place. Involve your child in the practical steps where possible, including removing access and reviewing subscriptions.

Address what might have drawn them to it

A child may be lonely, experiencing bullying, worried about being judged, or struggling to find someone available to listen. Others may simply have started through curiosity or for entertainment.

Help them identify trusted people they can approach, including someone outside the immediate family if this feels easier. Make regular space for conversation and activities they enjoy when you can. Where there is an underlying difficulty, work with the school or an appropriate professional to address it.

Look for changes in wellbeing

Pay attention if your child:

• Loses sleep or struggles to stop chatting with an AI.
• Withdraws from friends, family or activities.
• Becomes very distressed when the chatbot changes or is unavailable.
• Feels responsible for its apparent sadness, loneliness or jealousy.
• Increasingly relies on it for decisions or reassurance.
• Makes repeated or secret payments connected with access.

These changes warrant a supportive conversation about what is happening, including whether chatbot use is contributing to the difficulty.

Report harmful interactions

Stop an unsafe conversation and prioritise the child’s wellbeing. Record the service name, character or account details, dates and relevant messages where appropriate.

Report the issue to the provider and involve the school’s safeguarding lead or an appropriate professional where necessary. Do not download or share sexual images involving children, or ask a child to repeatedly revisit distressing material to prove what happened. If a sexualised image of a child is involved, you can read our guidance here.

What schools can do

SAIFCA recommends that schools:

• Teach pupils about the dangers in age-appropriate language. Cover emotional dependency, misleading affection, unreliable advice, sexual content, secrecy and privacy. Explain how human-like design can influence trust, and what pupils should do if a conversation becomes uncomfortable.
• Include these risks in staff training. Staff should recognise possible signs of attachment or harm, respond without shaming the child, and know when to involve the designated safeguarding lead.
• Record safeguarding concerns involving harmful chatbot interactions or apparent emotional dependency through the school’s established safeguarding procedures
• Give parents and carers practical information. Use brief guidance, newsletters or information sessions to explain what companions are, where children encounter them, warning signs and how to respond.
• Make clear in school policies that AI companion and relationship-simulating services are not permitted on school systems or recommended for pupils.
• Do not recommend AI companions as friends, confidants or substitutes for counselling or pastoral care in or out of school. Review school resources for features that encourage children to form emotional relationships with a system.
• Assess technology beyond its educational label. Consider personal conversations, simulated affection, data collection and whether a tool encourages repeated emotional reliance.
• Make support easy to find. Pupils should know which adults they can approach and how to ask for help, including when they find it difficult to explain what is wrong.
• Review filtering, monitoring and reporting arrangements. Include companion services where appropriate, while recognising that use outside school can also affect pupils.

Further information

What does the research tell us?

Evidence is growing, although research cannot yet tell us the full effects of years of AI companionship during childhood.

Several accessible sources help explain the concerns:

• Internet Matters’ Me, Myself and AI explores children’s use of chatbots, including for companionship, their trust in responses, and gaps in safeguards.
• Common Sense Media’s social AI companion assessments identified unacceptable risks for under-18s in the services assessed.
• UNICEF’s When AI becomes a friend explains concerns about children’s rights, emotional dependency and the need for stronger protection.

More specific experimental research also helps explain how design can influence children’s responses. A December 2025 preprint found that young adolescents rated a chatbot using relational language as more trustworthy and emotionally close than one using a more transparent, informational style, despite finding both similarly helpful.

SAIFCA’s recommendation reflects the combination of documented safety failures, emerging research and developmental concerns. We should not have to wait for children to experience lasting harm before introducing protective boundaries.

Have companies introduced new protections?

Some providers have restricted access or added safeguards:

• Character.AI began removing open-ended chat for under-18s in the United States in November 2025, with other markets to follow. Some creative features remained available.
• Meta announced a temporary global pause on teenagers’ access to its existing AI characters in January 2026 while developing an updated experience. Its general AI assistant remained available to teens.
• OpenAI introduced parental controls and age prediction, followed by its August 2026 announcement of ChatGPT for Teens with additional protections.

Australia’s eSafety Commissioner also reported changes following its transparency notices, including Character.AI removing under-18 chat access in Australia. Its March 2026 findings nevertheless identified serious earlier gaps, including failures by three services to refer users to crisis support when self-harm appeared in their prompts.

Company announcements describe intended or implemented changes; they do not independently demonstrate safety. Availability can also vary by country and account.

These developments do not change SAIFCA’s advice against AI companion use by children and teenagers.

What are governments doing?

Selected developments as of September 2026

These measures address different risks and are at different stages. Families should not assume that a service is safe simply because it is available or subject to regulation.

Children’s protection must come first

Families and schools have an important role, but they cannot carry this responsibility alone. Companies must be accountable for foreseeable harm, and governments must enforce protections that address emotional dependency as well as dangerous content.

SAIFCA supports beneficial applications of AI. We also believe children need space to develop their independence, judgement and relationships with people. Loneliness and gaps in access to care should be addressed through stronger human support, without normalising AI companions as the answer.

Read more in SAIFCA’s full guide to AI risks to children.