Effective 28 August 2026 · Last updated 28 August 2026

Responsible AI policy

Kioo puts a generative AI in the hands of six-year-olds. That deserves an honest explanation of what it does, what it refuses, and where it fails.

1. What the AI Tutor does

A pupil photographs a homework question. The tutor reads the image, works out what is being asked, and returns a step-by-step explanation in language a primary-school child can follow. The pupil can then ask follow-up questions in chat — "why did you divide by 2 there?" — for free within that session.

It is available on Kioo Pro, at roughly 30 photo questions a month. Everything else in Kioo — all 3,151 lessons and 9,451 quizzes — involves no AI at all.

2. Which models we use

Every answer is labelled as AI-generated in the app. We may change providers or models as better ones appear; we will keep this page current.

3. Guardrails

The tutor is deliberately narrow. It is not a general-purpose chatbot that happens to live in a school app.

• Scope-bounded prompting. System instructions confine the tutor to primary-school educational content — mathematics, science, English, social studies, religious education, literacy. It declines to engage outside that.
• Age-appropriate register. Answers are written for a child between roughly 6 and 13, in plain language, without sarcasm or adult framing.
• Refusals. The tutor refuses sexual, violent, hateful, self-harm, extremist and otherwise adult content, and refuses requests that are simply not schoolwork.
• Input filtering. Submitted images and text pass through automated safety filtering before reaching a model.
• No personal advice. The tutor does not give medical, legal, financial or relationship advice, and does not present itself as a friend, a counsellor or a person.
• No autonomous action. It produces explanations. It cannot change your account, spend money, contact anyone, or do anything outside answering the question in front of it.
• Rate limits. Credit limits and per-account rate limiting cap both cost and abuse.

No guardrail is perfect. Determined prompting can push any model off course. That is exactly why the reporting route below exists, and why the tutor sits behind an account rather than being open to the world.

4. What it gets wrong

We would rather say this out loud than let a child discover it in an exam:

• Arithmetic slips. Language models can produce a beautifully explained method with a wrong number in step three.
• Misread handwriting. Faint pencil, slanted writing, or several questions in one photo lead to the wrong question being answered.
• A different method. The tutor may solve something correctly using an approach the child's school has not taught yet. In the exam, the school's method is the one that counts.
• Curriculum drift. Answers are aligned to primary level, but they are not generated from the Uganda CBC syllabus document. Our lessons are curriculum-built; AI answers are not.
• Confident fabrication. On obscure factual questions a model may state something untrue with complete assurance.

Practical advice we give pupils and parents: ask the child to explain the working back to you. If they can, they have learned it. If they cannot, they have copied it.

5. Academic integrity

We designed the feature to teach rather than to launder homework:

• Answers lead with the method, not a bare final result.
• Follow-up questions are free while each new photo costs a credit — the pricing deliberately rewards understanding one question over racing through twenty.
• The tutor will not sit an assessment for a pupil, and we do not support its use in supervised examinations or assessed coursework where a school forbids it.

Ultimately this is between a pupil and their school. Kioo's job is to make the honest path the easier one.

6. What happens to the data

• The photo is transmitted over TLS, passed to the AI provider, and deleted from our servers once the answer is generated.
• The question and answer text are saved to your account so you can re-read the working. You can clear that history at any time from the AI tab.
• Content sent through these providers' commercial APIs is not used to train their models. They act as our processors and may only use it to return an answer.
• We do not use your questions to train any model of our own, and we do not sell them.

Full detail in the privacy policy.

7. Reporting a bad answer

Every AI response in the app carries a report control. Use it freely — for a wrong answer, a confusing one, or anything that does not belong in a children's app.

8. Human oversight

• Reported responses are reviewed by a person, not triaged by another model.
• Patterns in reports feed back into our system prompts and filtering.
• Where a safety failure is systematic rather than a one-off, we will disable the affected path until it is fixed, even if that means the tutor is offline.
• Reviewers see the minimum data needed to assess the report.

9. What we commit to

1. Every AI answer is labelled as AI-generated.
2. We will not quietly turn the tutor into a general chatbot or a companion.
3. We will not use children's questions to train models, ours or anyone else's.
4. We will not place advertising in AI answers.
5. We will name the providers we send content to, on this page, and update it when that changes.
6. We will say when we get it wrong, rather than describing a fault as a feature.

10. Contact

Questions about how Kioo uses AI: support@kioo.site. Safety concerns: safety@kioo.site.