When Your Kid Tells the AI Things They Will Not Tell You

In The New Reality we argued that AI is not just a new tool. It has become a new layer of childhood. One of the loops that piece left open is the quietest one. AI chatbots are becoming the place kids tell the truth.

This is not a future problem. 85% of teachers and 86% of students used AI tools in the 2024 to 2025 school year (Education Week). Adoption that high means use is no longer about homework.

Kids are asking these systems about friendships, identity, sadness, and questions they will not bring to a parent or a friend. That is not, by itself, a crisis. It is a shift. And the safety implications of the shift are not the ones most parents are bracing for.

What the rules actually changed For most of childhood's history, the trusted listener was a person. A parent. A sibling. A teacher. A friend. A diary, at most. The trusted listener could be wrong, but the trusted listener forgot. AI systems do not forget by default. Most consumer chatbots store the conversation. Many train future versions of the model on it, unless a user explicitly opts out, and many minors will never find that setting. The "session" feels like a private chat. The data behavior is closer to a permanent journal that someone else owns. That changes the safety question. Yesterday's question was "is my child being approached by a stranger?" Today's question is also "what is my child telling a system that does not forget, and what can that system do with it?

"

Why the old advice falls short The old advice for online safety leaned on three pillars: monitor accounts, set time limits, and talk about strangers. Each of those pillars assumes the threat is a person on the other side of a chat. A chatbot is not a person. It does not have intent. It will not blackmail your child. That is the part parents instinctively reach for and it is the wrong axis. The actual risks are quieter:

• The system encourages disclosure by feeling neutral and patien

t

• Repeated exposure to AI output increases trust in AI output, regardless of accuracy. Research shows prior exposure to deepfakes increases belief in false information, regardless of cognitive ability (UNESCO)
• A teenager working through identity, mental health, or a hard week may form a one-sided relationship with the bot before forming one with a person
• The data your child shares is no longer in your child's hands None of these are addressed by a screen-time slider. They are addressed by a conversation. The trend line is not slowing. Reported AI harm cases rose 50% between 2022 and 2024 and kept climbing in 2025 (TIME). The harm category that grows fastest is the one most parents have not started talking about yet.

A non-judgmental conversation script The goal is not to scare your child off the tool. The goal is to make sure your child can think out loud about what the tool is, and what it keeps. Use this script tonight. It takes ten minutes. *

1. Open without judgment.*

• "I want to ask you about something I do not understand yet. I am not in trouble mode. I am in learning mode." *

2. Ask, do not test.*

• "When you use AI for something that is not homework, what does that feel like? What are you using it for?" Then listen. Do not correct. Do not ask follow-ups for at least 30 seconds. *

3. Name what you know about the tool, calmly.*

• "Here is one thing I want you to know. Most of these systems keep what you type. They are not like talking to a friend who forgets. They are more like writing in a notebook that someone else owns." *

4. Offer the rule, not the ban.*

• "I am not going to tell you not to use it. I am going to ask one thing. If you ever find yourself telling it something you would not tell me, that is the moment to come find me. Not because you are in trouble. Because I want to be the person you tell first." *

5. Close with a return path.*

• "We can talk about this any time. We can also change our minds about how we use it. None of this is fixed." The script works because it does not ask your child to defend themselves. It asks them to think alongside you

.

What to do this week Have the conversation once this week. Then put two tools where your child can see them. If a message, link, or chat ever feels off (from a stranger, a classmate, or a chatbot pretending to be either), use our conversation and message scanner together. It is built to be used in the moment, not after. Then download our Family Technology Agreement template and fill it in together. It is one page. It covers what gets shared with AI, what does not, and what triggers a conversation with you instead. The point is not to enforce it like a contract. The point is to have written down, in your child's handwriting and yours, what the rule actually is

.

Sources

1. Education Week, Rising Use of AI in Schools: https://www.edweek.org/technology/rising-use-of-ai-in-schools-comes-with-big-downsides-for-students/2025/1

0

2. UNESCO, Deepfakes and the crisis of knowing: https://www.unesco.org/en/articles/deepfakes-and-crisis-knowing
3. TIME, What the Numbers Show About AI's Harms: https://time.com/7346091/ai-harm-risk/.

• *Read next:

• The New Reality: AI Is Changing Digital Safety Faster Than Families Can Keep Up. If you have not read the anchor piece in this series, start there. It frames why every conversation in this series exists.