No AI tutor is completely safe, and anyone who says so is overselling. Check whose account it runs on, whether answers can be wrong, where chats are stored and who can read them, what personal details it asks for, and how long sessions run. CoreMark Coach runs from a parent's Claude account, uses a first name only and works best with a parent nearby.
CoreMark makes CoreMark Coach, an AI tutor, so we have an interest in your answer. We describe what Coach is built to do and what it cannot promise, and we send you to Anthropic's own pages for anything about Claude's data practices.
Parents who ask this are usually thinking of three different worries: a stranger on the other end, wrong answers, and data going somewhere unknown. They need different checks, so we take them one at a time. None of what follows is legal or medical advice, and none of it replaces reading the terms of the service you choose.
Whose account does a child use?
Start with the age rules, because they decide who holds the account. Anthropic, which makes Claude, says: "We require all users to be at least 18 years old to create and use a Claude account" (Anthropic help centre). Its age-assurance page adds that it may ask users it believes are under 18 to verify their age. So CoreMark Coach is set up and run from a parent's Claude account, and we ask you never to let a child create their own.
Other services differ. OpenAI's terms set the minimum for ChatGPT at 13, with a parent or guardian's permission for anyone under 18, and Google's help centre sets 13 for personal Gemini accounts. Rules change, so read the current terms of whichever tool you use.
What happens when the AI is wrong?
Every AI tutor can state a wrong answer with confidence, which for a child is the most likely real harm. Coach is built to reduce it: it checks that each question's marks add up and that it sits in the Stage 9 curriculum, works out each answer independently before marking a child wrong, and runs diagrams through a checklist. If it finds its own mistake it says so and replaces the question.
Those are design intentions, not guarantees. The habit that protects your child is a weekly spot-check: pick one marked answer and compare it with a worked solution from a booster or textbook. If they disagree, ask Coach to show its working. The post on supervising a child using an AI tutor has the full routine.
Where do the chats go, and who can read them?
With Coach, the conversations live in the parent's own Claude account, under Anthropic's privacy terms, so you can look back through them. We do not run a separate system that stores your child's chats, and we make no claims beyond that about how Anthropic handles data. Read Anthropic's pages for yourself.
Two are worth knowing. On training, Anthropic's page says: "You choose to allow us to use your chats and coding sessions to improve Claude" (privacy.claude.com). On deletion, it says conversations you delete leave your history immediately and are removed from its back-end within 30 days (privacy.claude.com). Check your own account's settings, because options change.
What personal details should never be typed in?
Keep it to schoolwork. Coach is built to use a first name or nickname only, to stay on its own stage's schoolwork and never to ask for details beyond what tutoring needs. It does not ask for a child's school, address or photos of the child. The only pictures it asks for are of written work or drawings, such as a diagram your child drew on paper.
Make the same rule at home: no surname, school, phone number, address or photos of faces in a chat. If Coach ever asks for any of those, stop the session and write to us at [email protected].
How does Coach handle a child who is upset or stuck?
It is written to be warm and to treat mistakes as gaps to close rather than failures. If your child says a paper is too hard, Coach is designed to shorten it or switch to teaching, and its spec puts it plainly: kindness beats coverage. XP, levels and streaks reward turning up and fixing mistakes, and XP never goes down.
That is design, not a substitute for you. If a session ends in tears twice, stop and look at the cause. A tool cannot tell whether a child is anxious about exams in general, and a person can. Our piece on AI tutor vs human tutor in India covers when a person should come first.
What limits should you set?
- Session length. The Coach page suggests twenty to thirty minutes on most days works better than one long weekend session.
- Presence. Sit in for the first few sessions, then check in. You are the account holder and the adult.
- One topic per chat. Start a fresh chat for each session so you can look back through shorter transcripts.
- Off-topic chat. Coach redirects non-schoolwork requests, and you can too.
- Screens. Use the same rules you apply to other screen time.
What should you tell your child about using AI?
Say it in plain words before the first session. It is a tool that is good at explaining and sometimes wrong, so they should check anything that surprises them. It is not a person, so they do not need to be polite to it or afraid of it. They should not type anything about themselves beyond a first name, and they should tell you straight away if it says something odd or makes them uncomfortable.
Add one rule about effort: the point is to try first. A child who asks Coach for the answer before attempting the question gets less out of it, and Coach is built to ask them to try and to show working. Children usually respond to being treated as capable of understanding why.
What should you do if something goes wrong?
Stop the session, close the chat and keep the conversation so you can look at it. Decide whether the problem was a wrong answer, an unexpected reply or something about your child's reaction, because each calls for a different response. A wrong answer is fixed by checking with a worked solution. An odd or unsuitable reply is worth sending to us at [email protected] with what was said, so we can look into how Coach handled it. A distressed child needs you, not more tuning of the tool.
Which questions should you put to any AI tutor vendor?
Ask these of any product, including ours, and be wary of vague answers.
- Who must hold the account, and what age does the underlying service require?
- Where are conversations stored, and can I read and delete them?
- What personal details does it ask for, and can I avoid giving them?
- How does it behave when it is unsure or wrong?
- Is it affiliated with any exam board, and does it say so honestly?
- What is the price for the whole period, and does it renew automatically?
- Is there a person I can email, and how fast do they reply?
For Coach the answers are on the Coach page. Price is for the school year, with no auto-renewal, and the contact is [email protected].
What can we not promise?
That an AI will never be wrong, that a child will never see something unexpected, that memory features will always save, or that using any tool will raise marks. CoreMark is not affiliated with Anthropic or Cambridge, and Coach runs inside a service we do not control. Supervision is part of the design, not an add-on.
See how Coach works
CoreMark Coach is an AI tutor for Cambridge Lower Secondary Stage 9 that runs in Claude.
Frequently asked questions
We cannot call any AI tutor completely safe. Coach is built for children: a first name only, schoolwork only, a kind tone and checks on its own work. It runs from a parent's Claude account because Anthropic requires account holders to be 18 or over, and we recommend a parent sits in for the first few sessions.
No. Anthropic's help centre says it requires all users to be at least 18 to create and use a Claude account, and that it may ask users who appear to be under 18 to verify their age. Coach is meant to run from a parent's account with the parent involved.
Coach conversations sit in the parent's own Claude account under Anthropic's privacy terms, and CoreMark does not run a separate store of the chats. For what Anthropic does with data, read its privacy pages, which describe a user choice on training and deletion of conversations within 30 days.
Ask it to show its working and compare with a worked solution from a booster or textbook. Coach is built to re-derive answers and to say so and replace a question if it finds its own mistake, but no AI is perfect. A weekly spot-check of one marked answer is the habit that protects your child.
Sit in for the first few sessions, then check in regularly and read the end-of-paper report together. Because the account is yours, you can also look back through past chats. There is no fixed rule, and children who are anxious or easily distracted may need you nearby for longer.
Coach is built for Stage 9 students and we do not market it to under-13s. Anthropic's rules require adult account holders, and other services set their own minimum ages, for example 13 for ChatGPT and for personal Gemini accounts. Check the current terms of any service you consider.
Sources & further reading
- Anthropic, "Minimum age requirement", Claude help centre, support.claude.com
- Anthropic, "Age assurance on Claude", Claude help centre, support.claude.com
- Anthropic Privacy Center, "How do you use personal data in model training?", privacy.claude.com
- Anthropic Privacy Center, "Can you delete data sent via Claude?", privacy.claude.com
- OpenAI, Terms of Use, openai.com
- Google, Gemini Apps Help, age requirements, support.google.com
Last updated 20 October 2026. Details about other companies and research were checked against their own pages on 20 October 2026; check current terms, features and prices before you decide.
CoreMark Coach is an independent product that runs in Claude, made by Anthropic. CoreMark is not affiliated with Anthropic or with Cambridge Assessment International Education. AI can make mistakes: check important answers, and supervise your child's use.