⚠️ DRAFT — not yet in force. This document is published in draft, for transparency, while our organisation's registration is being completed. It has not yet been reviewed by a solicitor or a Data Protection Officer, and any detail shown as 'to be confirmed' is not yet finalised. We will replace it with a reviewed, signed version.

AI Tutor Notice — physmat.org

This Notice explains the optional AI tutor: that you are talking to an AI and not a person, what is sent to the AI model, where it is processed, what happens to what you type, and what we do if you tell it you are not safe. It is written to be read alongside our Privacy Policy (see §5), our Children's Privacy Notice, our Safeguarding Statement, and our Terms of Service. AI Act Art. 50 / ICO Std 4.

This feature is offered in the United Kingdom. It is not offered in the United States at this time.


Part 1 — The short version (shown the first time a child opens the tutor)

You're talking to an AI helper, not a person. - It is a computer helper, not a real teacher or a real person. - It can be wrong. Always check its answers yourself — don't just trust them. - It can see the problem you're working on and how much help you asked for. It does not know your name, your email, or your school. - What you type is saved in your account, so you can read it again. Your parent and your teacher cannot see it. - If you tell it that something is wrong or that you don't feel safe, a grown-up on our team can be told, so we can help. We won't just ignore it. OK, I understand — open the tutor.

(The wording above is the intended just-in-time message. The exact on-screen text, the age-appropriate reading level, and how the acknowledgement is recorded are.)


Part 2 — The fuller explanation

1. What the AI tutor is

The AI tutor is an optional, paid, opt-in feature. A parent or grown-up has to turn it on and pay for it before a child can use it — it is off until then. A child can use the whole platform — the problem bank, practice and grading — without the tutor.

This is a learning helper, not a person and not a marketing product. We do not advertise inside the tutor and we never send marketing through it.

2. You're talking to an AI, not a person

When a child opens the tutor, they are told — in language they can understand — that:

AI Act Art. 50: people must be told when they are interacting with an AI system. ICO Std 4: transparency for children. The exact on-screen presentation is.

3. What is sent to the AI model

To answer a question, the tutor sends the AI model only:

It does not send the child's name, email, parent's email, school, or account identifier. To the AI model the interaction is pseudonymous — it does not know who the child is.

We will not overstate this and we will not understate it: the four items above are what is sent, and the personal identifiers listed are not.

4. Where it is processed

The conversation is processed through our AI-inference provider — to be confirmed, depending on which model-inference path is configured. Either way, the provider receives only the pseudonymous payload described in §3. We intend this inference to run within the UK/EU region (eu-west-2). The full list of sub-processors, including both inference paths, is in the Privacy Policy §7.

5. What happens to what you type

6. If a child is not safe — safeguarding

If a child types something that suggests they may be at risk or not safe, we do not silently ignore it. We act under our Safeguarding Statement, which describes how concerns are handled and escalated to the people responsible for safeguarding.

We are building a system to detect such disclosures in the tutor and escalate them to a responsible person who can act.


7. Where to read more, and how to ask a question