The Danger Was Never Access

My son asked me why Godrej Fab was showing him ads on Youtube.

He’s on a supervised account, so everything he does online runs through settings I set up myself. Which made it a fair thing to ask. A detergent company was paying money to reach a boy who has never bought a box of detergent in his life, and he wanted to know why.

We got talking about how targeting works. I assumed I’d be the one explaining it. Instead he explained most of it to me, and then went somewhere I didn’t expect. He brought up surrogate advertising on his own. Alcohol can’t be advertised directly in India, he said, so the brands stick their name on soda and music and packaged water and advertise that instead. He’d noticed 100 Pipers and Black & White during the World Cup on ZEE5. He had the whole trick figured out.

I asked where he’d picked all this up. Not from me, and not from anyone at school. He’d asked Google’s Gemini one question, then a follow-up, then another, and it had answered each one as fast as he could type it.

What struck me later wasn’t the first answer. It was that the conversation never ran out of patience. He kept going. Why me? But I’m only eight. What if they know I’m eight? What if I tell you to buy it? None of those questions embarrassed the machine. It never told him he’d asked enough. It kept pace with his curiosity.

There was nothing here I could have blocked. He’d asked a legal question about the country he lives in, and he got clear, accurate answers. No filter should have stopped that, and none did. Everything worked the way it was built to work. That’s the part that stayed with me.

I went and looked at the parental controls anyway. There are two. You can switch the AI on, or you can switch it off. Off takes away the thing that has him learning faster than I did at his age. On leaves it all running and tells me almost nothing. Family Link let me know, once, on the first day, that he’d opened Gemini. Since then it’s kept a polite silence about everything he does in there.

There are other tools that go further. Bark reads his AI chats now and pings me if something in them crosses a line. Sensible sits inside the browser and watches every chatbot he opens. Both are well built, and I understand why a parent would pay for them. But they all rest on the same bet, which is to raise an alarm when there’s danger and stay quiet the rest of the time. They’re looking for the child in trouble, not the one who’s just curious. The old wall came down, and we’ve been busy pouring a new one on the same spot.

Access was never the real danger, though. That was an assumption we could afford when the world moved slower. Curiosity used to ration itself, and mostly it rationed itself against other people’s patience. A question cost you something. You had to find someone who knew the answer, catch them when they had a minute, and read their face to gauge how many more you could ask before you’d worn out your welcome. The one who answered twice and then sent you outside to play. A teacher already turning to the next child with her hand up. You learned to spend your questions carefully, because the patience in front of you was always visibly running down.

We called all of that friction and set out to remove it. But it wasn’t only friction. The cost was doing something. You don’t learn to weigh a question until asking one costs you something.

AI has taken that cost to almost nothing. That’s what’s actually changed in my house. It isn’t that my son can reach something harmful. It’s that he never has to wonder whether he’s asked one question too many.

For most of history, a parent’s job with information was to ration it. You decided what your child saw, and when, and how much. I learned it from my own father, who did it well and never once thought of it as a job. That job hasn’t gotten harder. It’s over. And whatever takes its place isn’t something a company can sell me, because it was never a product to begin with. The job was always mine.

I wrote in an earlier essay that judgment gets trained through failure, and that AI pushes its price up rather than removing it. This is that same argument, except now it’s standing in my kitchen. A boy who can reach everything has to grow into someone who can handle everything, and there’s no toggle for that anywhere in the settings. It comes slowly, across the years he still brings me a question before he needs the answer, and it comes from my being the person he chooses to bring it to. No agent installs that for me, and no alert can stand in for it. Either that work is already happening, quietly, or it isn’t happening at all.

So I haven’t built the agent. The tool would work well enough. But reaching for it first means answering the wrong question. The question was never how closely I can watch what he finds. It’s whether he still comes and tells me about it. That gets built now, in the small moments when he volunteers something I never asked about. The day he goes quiet, no dashboard is going to close the gap. It’ll just put a number on it.