How Kibu Uses AI

1. Our rules for AI

• No child talks to an AI. There is no chatbot on the child side of Kibu, and adding one would be a change we would ask your permission for.
• No AI provider may train on your family's data. We require this in writing.
• AI never makes a decision about your child on its own. It suggests; a person decides.
• We send the least we can. Where a pseudonym works, we send a pseudonym.
• Every AI feature is optional and separately revocable.
• AI-generated imagery is labelled as AI-generated.

2. Where we use it

We deliberately do not send your child's name, username, date of birth, photograph or account identifier to Anthropic. A pseudonym such as 'Child A' produces the same quality of suggestion, so there is no reason to send more.

3. The avatar, in detail

This is the only place a photograph of your child leaves Kibu, so we are specific.

Your child takes a photo with the camera — not from the photo library. It is sent once to Google's image model together with instructions to produce a friendly, fully-clothed, age-appropriate cartoon. Religious dress such as a hijab is preserved. The result is checked for safety before anyone sees it; if it fails, we fall back to a ready-made character rather than showing it.

The original photograph is then deleted. We do not store it. There is no way for us, or anyone else, to retrieve it later.

We do not perform facial recognition. We do not generate or store a faceprint, facial template, or any biometric measurement. We are not identifying anybody — we are drawing a picture. The photograph is used once and discarded.

The avatar is AI-generated. It is labelled as such in the app, and it is a stylised cartoon rather than a realistic likeness.

4. What we require of the AI companies

We verify these are in force rather than relying on published terms, and we re-check them at least annually. Our current providers and their status are at getkibu.com/subprocessors.

• They may not train their models on anything we send them.
• They may not keep it beyond what is needed to answer the request.
• They must delete it when we tell them to.
• They must tell us who their own sub-processors are.
• They must notify us of a breach.
• They must use approved transfer safeguards where data crosses a border.

5. What AI does not do at Kibu

• It does not talk to your child.
• It does not decide whether a mission is approved — you do.
• It does not decide whether an account is suspended — a person does.
• It does not profile your child for advertising, and does not rank or recommend content to them.
• It does not assess your child's character, ability or behaviour in any way that affects anything outside the app.

6. When it gets things wrong

AI output can be wrong. Mission suggestions are suggestions — read them before you use them. Weekly report commentary is a summary, not an assessment of your child.

Safety screening errs towards blocking. If we wrongly blocked something your child wrote, tell us and a person will review it.

If an avatar comes out wrong or upsetting, your child can regenerate it or pick a ready-made character instead.

7. If we change how we use AI

Adding an AI provider, or using an existing one for something new, means a new entry here, an update to our subprocessor list, and — where it involves your child — a fresh consent request. We will not extend an existing consent to cover a new purpose.