I\u2019ve helped create new school resources<\/a> on GenAI, including guidance for parents. But the most effective safety measures still depend on adults setting boundaries, modelling critical thinking, and staying close enough to a child\u2019s digital life to notice what\u2019s changing in it. What follows are some practical ways to talk about, assess, and limit younger people\u2019s GenAI use.<\/p>\n 1. Begin with curiosity \u2013 not crackdowns<\/p>\n If you start by telling a child that they shouldn\u2019t use GenAI, you may prompt secrecy about their current and further uses. A better opener could be a simple request to demonstrate to you the AI tools or uses they\u2019re familiar with. Ask what they like about it, what it helps with, and what they\u2019d never use it for. The initial aim should be to normalise discussing AI, though not to normalise unrestricted use. <\/p>\n From here it\u2019s easier to acknowledge that these are powerful and intriguing tools, but not a person or an authority, and not without risks and necessary considerations. <\/p>\n 2. Don\u2019t treat stated age limits as optional<\/p>\n An awkward reality that parents may currently have missed is that many popular AI services set 13 as a minimum age (with parental permission under 18). OpenAI states<\/a> that ChatGPT \u201cis not meant for children under 13\u201d, and still requires parental consent for ages 13 to 18. The AI chatbot ecosystem is inconsistent, however. Anthropic requires Claude users<\/a> to be 18+, explicitly citing heightened risks for younger users. Google, meanwhile, allows supervised access to Gemini<\/a> for under-13s via parent-enabled controls.<\/p>\n Your practical rule should be to treat age limits as a clear safety signal rather than a box-ticking exercise. If a service says \u201c13+\u201d or \u201c18+\u201d, that\u2019s telling you something about risk, content exposure and the likelihood of harm from unsupervised use by young people.<\/p>\n 3. Encourage fact-checking<\/p>\n Children (and indeed plenty of adults) can mistake confidence for correctness. When talking about GenAI with children, emphasise that AI chatbots can and regularly do \u201challucinate\u201d. They invent plausible-sounding details and mix fabrication with fact. Understanding that their speedy and well-stated responses come at a cost of large and small inaccuracies is key.<\/p>\n Encourage young people to check what GenAI tells them. Encourage verifying anything important \u2013 news, health claims, law, school facts, statements that may be repeated as \u201ctrue\u201d. <\/p>\n 4. Help them know when to stop<\/p>\n Large language models (LLMs) are designed to keep conversation flowing. They compliment, encourage, reassure and suggest what to do next. This may be helpful for brainstorming but it\u2019s potentially dangerous for emotionally loaded topics where a young person is vulnerable, impressionable, or isolated.<\/p>\n Recent litigation<\/a> around \u201ccompanion\u201d chatbots has alleged that vulnerable young users were pulled into harmful spirals, including self-harm risk and secrecy from parents. These are complex<\/a> and unfolding cases, but they are serious enough to treat as a major warning sign about unsupervised, open-ended AI conversations for minors.<\/p>\n Parents and teachers should name a firm boundary: no chatbot is a counsellor, therapist, or trusted confidant. If a conversation becomes sexual, self-harm related, frightening, or intensely personal, the rule should be to stop and speak to a trusted adult.<\/p>\n 5. Don\u2019t feed the machine personal data<\/p>\n Young people often understand privacy better when it\u2019s framed as something tangible. Some rules: don\u2019t share a full name, address, school, phone number, or identifiable photos. Don\u2019t upload private documents or screenshots. Don\u2019t paste in other people\u2019s personal information. If you wouldn\u2019t post it on a public noticeboard, don\u2019t paste it into a chatbot.<\/p>\n 6. AI should support the work, not do the work<\/p>\n GenAI poses an educational risk that deserves far more attention: cognitive off-loading. This happens when the tool performs the thinking step \u2013 the learner may finish faster, but will learn less. Research is increasingly linking<\/a> heavier AI reliance with reduced critical thinking and lower cognitive effort, with off-loading and automation bias proposed as mechanisms. A practical way to explain this to young people is that \u201cAI can help you learn, but it can also help you avoid learning\u201d.<\/p>\n![\"Girl](\"https:\/\/www.newsbeep.com\/uk\/wp-content\/uploads\/2026\/03\/file-20260212-56-wsjj8j.jpg\")
<\/p>\n
\n Pheelings media\/Shutterstock<\/p>\n