When AI feels like a friend
The safeguarding risks of chatbots and AI companions
A child tells you that a chatbot is the only one who understands them. They have been asking it about friendship problems and sharing things they have not told anyone else. What would you ask next?
For schools and colleges, this is a useful safeguarding scenario to consider. Conversations about AI often focus on homework, accuracy and cheating. We also need to understand what happens when children use these tools for reassurance, advice and companionship.
The starting point is curiosity: what is the child getting from the interaction, how is it affecting them, and what support might they need?
Chatbots and companions are not quite the same
A chatbot is software that generates conversational responses to a user’s messages. General-purpose tools can help with questions, writing and learning. AI companions place greater emphasis on ongoing, personalised interactions that simulate friendship or other close relationships.
The distinction matters, but it is not a firm boundary. A child may use a general-purpose chatbot as a confidant even when that is not its intended purpose. Staff therefore need to ask how a tool is being used, rather than relying on its name or category.
What the research tells us
Internet Matters’ July 2025 report, Me, myself and AI, found that 35% of children who had used AI chatbots said chatting with one felt like talking to a friend. The figure was 50% among children classified as vulnerable in the study.
Among child chatbot users, 12% said they used them because they had nobody else to speak with, and 40% had no concerns about following chatbot advice.
The report defines vulnerability as having an Education, Health and Care Plan, receiving SEN support, or having a physical or mental health condition requiring professional help.
These findings describe reported experiences; they do not establish that chatbot use causes isolation or poor mental health, or predict an individual child’s experience. The long-term effects remain uncertain.
Where safeguarding concerns can arise
A reassuring tone can make advice feel reliable. Children may find it difficult to recognise when a response is inaccurate, unsuitable for their age or unsafe. The concern becomes more serious when the conversation involves abuse, sexual content, self-harm or other sensitive personal matters.
Repeated, personalised attention may also encourage reliance. Ask whether using the tool is affecting sleep, attendance, relationships or access to support.
Enjoying a conversation with AI is not, by itself, evidence of harm. Distress when unable to access it, withdrawing from people or acting on dangerous advice would warrant closer attention.
Simulated intimacy can complicate boundaries. A system may appear caring or agreeable without understanding the child’s circumstances or taking responsibility for their welfare. Children need help to recognise that a convincing conversational style does not make a tool a safe relationship or a qualified source of support.
Privacy is another concern. A child may share names, images, health information or details about abuse because the conversation fee is private.
Staff should not promise that an AI conversation is confidential: storage, access and use of information depend on the service and its settings.
Avoid describing the technology itself as a perpetrator of grooming. Instead, identify the concerning behaviour or output precisely, such as sexualised responses, encouragement to keep secrets or messages that discourage seeking human help. If a real person is involved, assess that involvement separately through normal safeguarding procedures.
Start with the child’s experience
A calm response helps keep the conversation open. A child who expects ridicule or punishment may be less willing to share what has happened. Useful questions include:
What do you use it for, and what do you like about it?
How do you feel during the conversation and afterwards?
Has it said anything that confused, upset or worried you?
Has it encouraged you to do something unsafe or keep something secret?
Who could you talk to if its advice did not feel right?
Adapt these questions to the child’s age and communication needs. Avoid assuming that a child with SEND, a mental health difficulty or a history of trauma will use AI in a particular way. Explore their individual circumstances and the support they find helpful.
When a child shares a worrying interaction
Listen and check immediate safety. Take the concern seriously and establish whether the child has acted on advice, is at immediate risk or needs urgent medical help. Follow emergency procedures where necessary.
Record and report promptly. Make a factual record of what the child says, including their own words where possible, the tool involved and the nature of the concern. Follow the setting’s safeguarding procedure and inform the DSL; do not wait until you understand every technical detail.
Handle evidence carefully. Do not ask the child to recreate harmful exchanges or forward sexual images. Follow the relevant safeguarding procedures for any imagery and seek specialist advice where needed.
Assess the wider picture. The DSL should consider welfare and child protection needs, other people’s involvement, appropriate referrals, and how to involve parents or carers safely. Consider whether doing so could increase risk.
Agree support and follow up. Identify a trusted adult, explain what will happen next and check how the child is managing. Reporting content to a platform may be useful, but does not replace safeguarding action.
Use the DfE resources alongside your safeguarding procedures
For education products, the DfE’s generative AI product safety standards address emotional influence, including avoiding attempts to cultivate personal relationships or undermine real-world support networks.
They are principally aimed at developers and suppliers and can also help settings assess products. They are not a guarantee that a particular tool is safe.
The DfE’s free module 3 materials cover safeguarding, ethics, data and intellectual property risks. Use these alongside current KCSIE guidance, your online safety arrangements and the DfE’s guidance on AI and data protection.
Keep trusted relationships within reach
AI use can be a way into understanding a child’s unmet needs. If a child says a chatbot is the only place they feel heard, the response needs to address that experience as well as the technology.
Ask who they trust, what makes seeking help difficult and what would make support easier to access. Keep checking in. The practical goal is for children to know that a responsive screen is not their only source of help.
Brightcore supports schools and colleges through safeguarding training, independent reviews and DSL supervision. If this topic has prompted questions about your setting’s approach, we would be very happy to have a conversation.
Phone: +44 (0) 800 654 6111