Kosher AI for kids - a parent's setup guide

Kosher AI for kids depends less on the software than on the ten minutes a parent spends setting it up. A filtered tool with nothing configured does less than it could, and a well-configured one still needs a parent the child will talk to. This guide walks through setting up a child's account in Navōn, what each control actually does, and the conversation worth having before your child starts.

If you are still choosing a tool, start with Safe AI for kids: what a parent should actually check, which covers how to test any product. This piece assumes you have chosen, and are now setting it up.

Step one: give your child their own account

The controls attach to a child's account, not to a device, so each child needs an account of their own. Letting a child use yours means they are held to your settings, which are an adult's.

In Settings, open the Family tab, which is the Family Hub. There are two ways to add a child.

• Add them as a family member. You create the child's login yourself, either a username and password or an email address that receives an invitation, and choose their safety level, tools and time limits in the same few steps.
• Link an account they already have, through school for example. Enter your child's email and they receive a secure link. Once they accept, the account is linked to yours and you can set controls on it.

Step two: the controls, one by one

Every control is set per child, so a ten-year-old and a fifteen-year-old can be treated differently, as they should be.

Safety level

Three settings: Strict, General and Minimal. General is the default for a child's account. Strict takes the narrowest reading of every subject. Minimal leaves only the platform's own rules, which are never off for anyone. For a younger child, Strict is a sensible place to start. It is easier to relax a setting later than to tighten one a child has got used to.

Blocked topics

Your own list, on top of everything Navōn already filters. Write a few words for each subject, the way it would actually come up: "horror films", "dieting". The list applies to every tool, so a topic closed in chat is closed in the Image tool too. When a request touches one, your child is told plainly that it is a topic their parent or school has asked Navōn not to discuss.

The list is matched against the words of the request. If your child writes in Hebrew or Yiddish, add the word in that language as well.

One exception is deliberate. If a child writes about being hurt, or about hurting themselves, a blocked topic does not refuse them. A parent who blocks a word like "abuse" means keep the subject away from my child, not turn my child away when they are in trouble, so a message that is a real disclosure gets a warm answer that points them to an adult they trust.

Tools

• Research: looking things up and reading sources, which is the News area.
• Homework help: the study tools in the Classroom, including the AI Tutor, the Study Helper and Torah Studies.
• Creative tools: making images, video and stories. Off is a reasonable choice for a younger child.
• Approve submissions: you see schoolwork before it is handed in to a class.

The first three are checked on Navōn's servers on every request, not just hidden from the menu. With homework help off, for instance, your child is told the homework and study helpers have been turned off, and the rest of Navōn still works.

Daily time and hours

A daily budget in minutes, and a window of hours when Navōn may be used at all. The hours are checked against your child's own timezone. Time is counted as the minutes in which your child actually made a request, not the minutes a tab sat open, so a budget of 30 minutes means 30 minutes of use. Outside the window, your child sees the hours and that their parent or school set them.

If your child's school also sets blocked topics, a safety level or hours, the two are combined and the stricter wins on each. You cannot switch off something the school has blocked, and the school cannot loosen something you have set.

Alerts when your child writes about distress

This is the control worth leaving on, and it is on unless you turn it off. If Navōn believes your child has written something about being hurt, or about hurting themselves, you get an email within moments rather than a summary at the end of the week. It quotes a short fragment, enough to know it is real, and links to the Family Hub, where every alert about that child is kept.

Three things to know. Your child is not told that you were emailed, and still gets a warm reply pointing them to an adult they trust. Alerts are limited to one per child per hour, so a hard evening produces one email, not five. And Navōn is a program that can be wrong in both directions: it can miss something said indirectly, and it can alert on something that was not serious. Treat an alert as a reason to go and sit with your child, not as a diagnosis.

What you can see, and how much to look

The Family Hub keeps a history of what your child asked and what Navōn answered, with refusals marked. It is there so you can understand a pattern, not so you read every message. A teenager who learns that every question is read tends to stop asking honest ones, which is the opposite of what you want from a tool you chose because you could trust it.

A reasonable rhythm is to look when something prompts you to: an alert, a refused request, or a change in your child you want to understand.

Step three: why kosher AI for kids starts with a conversation

No setting replaces this, and it takes about as long as the setup.

1. Tell them plainly what you can see: that their history is visible to you, and that you will be told if they write something that sounds like they are hurting. Children cope far better with rules they know than with rules they discover.
2. Explain that it can be wrong. It sounds equally sure of itself whether it is right or not. Anything that matters, a fact for a project, a date, anything about halacha, gets checked in a sefer or with a teacher.
3. Say what it is not. It is not a rav, and a sheilah goes to a rav or to you. It is not a friend either, however warmly it writes.
4. Agree on what happens when something odd comes back. The best outcome is that they come and show you, so say in advance that they will not be in trouble for it.
5. Sit with them the first few times. What they ask is interesting, and it tells you more than any report.

Then look at the settings again every few months. A setting that suited a ten-year-old will start to chafe at twelve, and a child who has outgrown a rule is better off having it changed with you than worked around without you.