How to talk to your kids about AI

My 12-year-old came to me wanting to talk. He had seen the news about alignment problems in AI models and what it might mean to push frontier systems out without guardrails, and he had questions. Not panic, questions. I work in this field, and I still didn’t have a ready answer, because the honest one is complicated and the reassuring one is a lie.

Most parents will have this conversation in the next year or two, if they haven’t already, and it may not arrive as fear. It may arrive as a kid who has been paying attention and wants to be taken seriously.

Find out what he actually wants to know

A kid who raises AI is not always frightened. Sometimes he is curious, sometimes he wants to test whether you know more than the news anchor, sometimes he has noticed that adults look worried about something invisible and wants to know what a kid is supposed to do about it. Those are different conversations.

So ask before you explain. What did they say? What do you think it means? Which part stuck with you? My son’s questions were specific: what does it mean for a model to be misaligned, and why would anyone release something powerful without the safety work done first. Those deserve real answers, not a pat on the head. If instead the worry turns out to be that a job he wanted might disappear, or that you might lose yours, answer that one.

Don’t say everything is fine

The temptation is to close the subject fast. “Don’t worry, it’s just the news.” Kids know when they’re being managed, and a 12-year-old especially. If he later reads that serious researchers do worry about this, you have spent your credibility on one comfortable sentence.

What works better is the version adults use with each other. Some smart people are worried about real risks. A lot of those same people work on this full time, and making sure it goes well is their whole job. It is not your job to carry this.

Children handle “this is a real problem that grown-ups are working on” much better than “there’s nothing to see.” The first respects them. The second tells them you either don’t know or won’t say, and they will go find the answer somewhere with worse intentions than you.

Demystify before you reassure

Nothing lowers fear like knowing how the thing works. A chatbot stops being a mind in a box once your kid learns it is predicting the next likely word from an enormous pile of text, that it forgets him the moment the conversation ends unless the app saves things, and that it doesn’t want anything the way he wants a snack.

You don’t need to be technical. A few explanations that land with kids:

It’s a very good autocomplete. It has read more than any human could and guesses what should come next. That is why it sounds smart and why it can be confidently wrong.

It doesn’t know you. Each conversation starts fresh unless the app stores it, and storing it is the app’s decision, not the AI’s.

It has no body and no plans for its afternoon. It runs when someone asks it something and stops when they leave.

Then let him poke at it. Sit together and ask it hard questions. Ask it something you both know the answer to and watch it stumble. Ask it to explain a joke. Ask it whether it wants to take over the world and read the answer together, then ask it how you would know if it were lying. That last one is the alignment problem in a form a 12-year-old can hold, and the answer (you can’t, from the outside, which is why people test these systems so hard) is more useful than any reassurance.

Tell the truth about the risks, at the right size

He will ask directly: so could it hurt people? The honest answer, sized for a child, goes something like this.

Not the way the movies show. The real worries are that people use AI to trick other people with fake videos, fake voices and scams; that it can be wrong and people believe it anyway; and that far more powerful future versions might do things nobody intended, especially if companies race to ship them before the safety work is finished. That last one is what my son had read about, and it is real. It is also the reason labs publish their own test results, including the embarrassing ones, and the reason people argue in public about rules for releasing these systems.

Two of those worries are things a kid can do something about, which matters more than anything else in this article.

Give him something to do

Whatever fear there is usually comes from helplessness. A child who knows how to check whether a video is real, who knows the voice on the phone claiming to be grandma might not be grandma, who asks “how do you know that?” when the chatbot makes a claim, is no longer a bystander. He is a participant.

What to teach depends on age.

Younger children, roughly 6 to 9, need two ideas: AI is a tool, like a calculator that talks, and it is not a person and not a friend. Keep them out of open chatbots entirely, or use something built for kids where a grown-up can see the conversations.

Preteens, 10 to 13, can learn to fact check. Show them how to confirm a claim with a second source, how to spot a generated image (hands, text and reflections still give it away, though less every year), and why they should never give a chatbot personal details. This is also the age to talk about deepfakes plainly, because they will run into one.

Teenagers can handle the real debate. Read one alarming article together and one skeptical one. Ask which arguments hold up. Teenagers respond well to being treated as capable of weighing evidence, and this is a topic where the evidence genuinely cuts both ways.

Model the relationship you want him to have

Kids copy how you use technology far more than they listen to what you say about it. Treat AI as an oracle and they will. Ask it things and then check, argue with it, say “that’s wrong” out loud when it’s wrong, and they learn that instead.

Let him watch you use it for something useful and dull, like drafting an email or working out what a form is asking for. The more ordinary it looks, the less it looks like a threat.

Leave the door open

One conversation won’t do it. The news will keep coming and the questions will come back in new shapes. Make it clear he can raise it any time, and that “I don’t know” is an answer you’re allowed to give. Say it when it’s true. You are not trying to be the expert. You are trying to be the person he comes to when he wants to understand something, and that only works if he trusts you to tell him the truth.

If you want a way to end the first conversation, try this: say the true things (yes, it’s real, some people are worried, a lot of people are working on it, and nothing about it changes what happens in this house tonight), then open a chatbot together and ask it for something silly. A poem about the cat. A rap about homework. It will probably be bad. He will probably laugh. A kid who can laugh at the thing and also ask serious questions about it is exactly where you want him.