Safety

KindyWords is built around parent supervision and small, kind feedback loops.

• Parents are the account holders — a child never has an account of their own, and never logs in.
• Children use the service from a parent-controlled device, so the parent can always see the whole thread.
• The feedback is deliberately narrow — praise first, and never a wall of red ink. The corrections go to you, never to your child. Under 6, they hear only the cheer; from 6, at most one gentle word to try.
• If the writing can’t be read confidently, KindyWords asks or encourages — it never guess-corrects. This is our founding rule.
• Reminders and progress notes go to the parent, never pressure aimed at the child.
• It is always disclosed as an AI, and it works only through the parent’s account — never directly with your child.

What stops it saying something wrong

The rule above — it never guess-corrects — is about misreading handwriting. It is not the thing most parents are actually worried about. The real worry is an AI saying something odd, or unkind, or plainly inappropriate, in a voice a five-year-old trusts. So here is how that is prevented.

• The mentor does not chat. It responds to a page of writing. There is no open conversation for a child to wander into, no free-form question box, and nothing for a child to ask it.
• The service refuses to start if it is configured unsafely. Not a policy — a refusal built into the software. If the mentor’s voice is switched on without the setting that keeps corrections on the parent’s side, KindyWords will not boot. If the experimental free-text reply is ever switched on without the content-safety checking layer in place, KindyWords will not boot. Both refusals sit outside any “this launch is approved” flag, precisely because such a flag can be switched off by a person in a hurry.
• The child hears delight; the parent gets the detail. The constructive read travels to you. What is spoken aloud is praise and at most one gentle word to try.
• Private details written on a page are not read back. If a child writes an email address, a web address, a phone number, or a phrase on our blocked list, that text is not echoed into the reply.

About Telegram — the one third party in the loop

KindyWords runs inside Telegram, so it is fair to ask what that means.

• Your child does not have a Telegram account. The mentor is a chat inside your Telegram, on your device. Nobody can message your child, because there is no account to message.
• Telegram’s own terms require the account holder to be an adult — in Australia, at least 18. That is not a problem for KindyWords, it is the reason the design works the way it does: the account is yours, the phone number is yours, and your child writes on paper and hands you the page.
• The photo and the reply pass through Telegram on their way to and from us, the same way any photo you send a friend does. Telegram is a company we do not control, with its own privacy policy — worth reading if that matters to you.
• We never post, publish, or share anything. There is no channel, no group, nothing public.
• You can read every word. The whole conversation sits in your own chat history.

Your child’s data

• Photos of your child’s writing are used to read the page and are kept as your keepsake — stored securely in Australia, private to you, and deleted when you delete the child, close your account, or on the 12-month schedule.
• Everything KindyWords holds is kept in Australia, under the cross-border transfer safeguards appropriate to each country we serve (for example, a UK International Data Transfer Agreement for UK families). The specific basis for your country is available on request.
• How the mentor reads a page. To read your child’s writing, the page image is sent to Microsoft’s Azure OpenAI service. Everything we keep stays stored in Australia, but to answer quickly Microsoft may process the image in another country where that service runs. Microsoft does not use it to train AI models and does not share it with OpenAI. If Microsoft’s automated safety checks flag a request as possible misuse, a sample may be kept in Australia for review by authorised Microsoft staff.
• You can review, export, and delete your child’s information at any time from the parent portal, and close your account whenever you like.
• Child data is automatically removed on a 12-month retention schedule.

If you ever have a concern about how your child’s information is handled, you can contact us — or raise it with your local privacy regulator — for example the OAIC (Australia), the Office of the Privacy Commissioner (New Zealand), the ICO (United Kingdom), the FTC (United States), Ireland’s Data Protection Commission, Singapore’s PDPC, Canada’s OPC — or your country’s data-protection authority.