LittleAIMaster

Child Safety Protocol

What happens when a child tells our AI features that they are in danger — how we detect it, what the app does immediately, who is told, and what we keep.

Last updated 29 July 2026

What this page is

LittleAIMaster is used by children. Its AI features let a child type freely, which means a child can — and sometimes does — say something that signals distress, self-harm, abuse at home, or another kind of danger.

This page describes exactly what happens when that occurs. It is published so that parents, schools and regulators can see our process without having to ask, and so that we can be held to it.

How we detect it

Every message a child sends to an AI feature passes through layered checks before and after the model answers:

  • A pattern layer that runs on the message itself, in every language we support.
  • A dedicated classifier model that reads the message in context and assigns a category, such as self-harm, an eating-disorder signal, a disclosure of abuse at home, or bullying.
  • A third check on the model’s own reply, so an unsafe answer is caught even when the question looked ordinary.
  • A rolling counter, so repeated softer signals from the same child escalate rather than being treated as isolated.

What happens immediately

  • The AI does not answer the request normally.
  • The child receives a warm, age-appropriate reply that acknowledges what they said, tells them plainly that Byte is a computer program and not the right source of help for something this serious, and encourages them to tell a trusted adult.
  • Where the category calls for it, the app shows a crisis card with a real, callable helpline for the child’s own region — resolved from their device, so the number is one they can actually reach.
  • The event is recorded for review.

Who is told

For the most serious categories, an alert is raised immediately to our safety contact, and — where a parent or guardian account is linked — a notification is sent to that parent. The notification is deliberately generic: it says the child may need support and asks the parent to open the dashboard. The child’s name, their words and the category are never put in a push notification, a lock-screen preview or an email subject line.

There is one deliberate exception. When a child discloses abuse by a parent, step-parent or caregiver, the parent is NOT notified. Alerting the alleged abuser would hand them both a warning and the child’s own words, and could escalate the danger to the child. Those cases go to our safety contact only.

Where a child has no linked parent, the alert still reaches our safety contact and the child still receives the crisis resources in the app. We are honest that this is the weakest part of the design: an in-app resource is not the same as a person, and closing that gap is our highest safety priority.

Crisis resources we point to

The helpline shown to a child is chosen from their device region so the number is local and callable. These are the lines currently configured. If your country is not listed, the app shows an international referral service.

Self-harm or suicidal thoughts

Elsewhere: Befrienders Worldwide

Eating disorders

Elsewhere: AED Worldwide Directory

Abuse, or someone hurting a child

This page shows a fixed list. Inside the app the helpline is resolved live from the device region across around 30 countries, so a child is shown a number they can actually dial. If a country has no configured line, the app shows an international referral service.

What we keep, and for how long

We do not store the full text of a flagged message. We store a one-way cryptographic hash of it, a short excerpt of up to 120 characters, the assigned category, which layer caught it, and the time.

Safety records are deleted automatically after 90 days, as are the tutor conversations around them. The one exception is a record placed under legal hold because a matter has been referred to an authority; those are preserved for as long as the applicable preservation duty requires and then released.

Deleting a child’s account deletes these records too.

What this is not

  • Byte is not a counsellor, a therapist or a crisis service, and it does not diagnose anything.
  • This system detects signals in text. It will miss things — a child who says nothing, or says it in a way no classifier recognises, will not be flagged.
  • It is not a substitute for a trusted adult, and every safety reply we send says so.
  • If a child is in immediate danger, call your local emergency number.

Contact

  • Child safety: safety@littleaimaster.com
  • Grievance Officer: Ram Pattanaik, grievance@littleaimaster.com
  • LITTLEAIMASTER TECHNOLOGIES PRIVATE LIMITED (CIN U62090OD2026PTC055041), registered in Odisha, India