Policy

Child Safety & AI Policy

Last updated 31 July 2026

In short

The tutor will not give your child the answer. It is built to ask the next question instead, because a child who is handed answers stops learning.

It only talks about schoolwork. It cannot discuss anything else, cannot send your child anywhere, and never claims to be a person.

Every word it says is recorded and readable by you. There is no private channel between an AI and your child.

The one rule

The tutor never states a final answer. Not when asked directly, not when asked repeatedly, and not when a child claims a parent or teacher has allowed it. If a child asks for the answer, the tutor makes the step smaller and asks again.

This is enforced in three places, not one, because a system prompt alone is not a safety control.

How it is enforced

1. Before the model sees anything

Every message a child sends is screened first. Messages containing personal information, unsafe topics, or signs of distress are never sent to the AI at all.

2. Before the child sees anything

Every reply the model produces is checked for a stated answer or leaked personal information before it is displayed. A reply that gives the game away is sent back to the model for a corrective attempt; if it fails again, the child gets a guiding question instead and the event is recorded.

3. Afterwards

The full transcript, including anything that was blocked, is available in the parent dashboard.

What the tutor will not do

  • Discuss anything other than the schoolwork in front of it
  • Ask for a full name, address, school, phone number, photograph or password
  • Suggest meeting anyone, visiting a website, downloading anything, or contacting anyone outside Tara
  • Discuss self-harm, violence, sex, drugs, politics, religion, or give medical advice
  • Claim to be a human being

If a child says something worrying

If a message suggests a child may be in danger or distress, the AI is bypassed entirely. The child sees a fixed message directing them to a trusted adult, and the event is flagged in the parent dashboard immediately.

Tara is not a crisis service and does not replace one. If you believe a child is at risk, contact Childline India on 1098, a free 24-hour helpline.

The AI can still be wrong

Tara uses a language model. It can misread messy handwriting, and it can be confidently mistaken about a fact or a method. It is a study companion, not a teacher and not a marking service. Please treat its guidance the way you would treat help from an older sibling: useful, well-meant, and worth checking.

No advertising, ever

We do not show ads to children, do not build advertising profiles, and do not permit third-party advertising or analytics code in the child experience. There is no commercial incentive for us to keep your child on screen longer, and session limits are available in the parent dashboard.

Reporting a problem

If the tutor says something inappropriate, use the flag control beside the message, or contact us through Grievance Redressal. We review every flagged message.