I saw this first hand when I visited Hamilton County, Tennessee; a district that takes STEM seriously, and had invited me out to observe its computer science and maker program. I tried to access ChatGPT on the school wifi, but quickly realized it was blocked. The fix was easy: I simply used the mobile data on my phone to access it instead. Every student in that building would have done the same. This is not a district that’s behind or an outlier. It is doing what the majority of over 13,000 school districts in this country have defaulted to.

It reminded me of a story an English teacher told me, about her efforts to stop her class using AI for homework. While marking her students’ assignments, she noticed about a quarter of the essays she was reading looked remarkably similar. She sent an email: ‘If you used AI, please let me know.’ A hundred per cent of the students responded admitting that they had.  Her policy hadn’t succeeded at anything, other than to show her that children aren’t hiding AI from us. Instead, we are hiding from AI, which is costing us the one conversation that would help them.

I’ve been to school districts across the US over the last three years, and I realized I was seeing repeats of this same story: a blanket block failing to stop the use of a technology that has run rampant, and instead only succeeding in moving it out of sight from the adults who should be supervising it. More than a third of entry-level jobs require AI skills, according to Nace’s 2026 Job Outlook Survey. Ultimately, every district that blocks AI is now making a decision about who gets hired in 2030, and simultaneously pretending it isn’t a decision at all.

AI needs to be built with children in mind

The majority of conversations around AI in education usually focus on whether kids might use AI to cheat in an essay. But the stakes are far higher than that: AI is shaping how we all think, and for children in their formative years, it’s shaping how they learn, socialize, and make decisions. The problem is that almost none of the development driving these systems is treating children’s safety as a priority, nor are they building it in a way that allows kids to learn and use it productively. It would seem, therefore, that we’re running a social experiment on millions of kids without knowing the long-term effect.

The people who study young minds are split. In an NBC survey of the American Psychiatric Association and American Counseling Association, 86% of psychiatrists agreed that AI use among teens would inhibit brain development, replacing human interaction with lower-quality synthetic socialisation. However, 64% agreed that children would learn effectively from AI platforms that tailor lessons in a way that those teaching a class of thirty rarely can. Both can be true, but which outcome we get depends on the tools our children can access, and whether anyone is teaching them to use AI productively. 

The case for restriction might be strong for social media, but it is weak for AI. Feeds hold a child’s attention, but there is no real reason for a child to be fluent in it. Meanwhile, AI packages its harm and skill in the same product. When we impose blocks that are easily surpassed by VPNs or a Wi-Fi switch, we let AI developers off the hook for children’s safety, and schools to step back from the conversation entirely. Out of sight, out of mind. But this isn’t a sustainable plan. 

Schools are struggling to keep up

Today, four out of five students are using AI for schoolwork, according to Stanford’s AI Index. Despite this, only 6% of teachers say their school’s AI policy is clear, and even then, few school districts have worked out a plan for making the transition from a teacher using AI for themselves to using it with thirty or so students. 

When you consider the pace of AI’s acceleration, it’s difficult to point fingers. Within two years of ChatGPT’s release, roughly 40% of US adults had tried it; the PC took twelve years. In the years since AI has been developing, global K-12 edtech funding fell by 82%. Schools were asked to absorb the fastest technology shift in modern history with less support than ever. So, school districts are carrying the liability on their own backs: one superintendent I spoke with recently told me his policy is simple – ‘if it didn’t come from us, it’s on you’. A fairly understandable way to manage risk, but ultimately a terrible way to manage a technology moving so fast, as it leaves the teacher the least resourced to act.

Meanwhile, the gap is widening

Handle it well, and AI could be the greatest equalizer education has ever had. Handle it badly, and we will watch it deepen every single divide we already have. Over 85% of school administrators consider AI education valuable, according to the College Board. Yet, 45% are reporting that access is restricted within their classrooms. The kids who belong to school districts where the only AI policy is no AI will one day be competing for the same entry-level jobs as others who have become literate during their same school years.

During my classroom visits across the country, I see both sides of the coin day-to-day. In some districts, the only AI you see is in an English department’s detection software designed to catch students cheating. In a district right next door, I’ve seen leaders run AI summits for educators, administrators, and AI innovators to learn first-hand how to use and develop frontier technologies so they can best benefit their students. Two districts can be so close in geography and yet headed in completely opposite trajectories. Schools trying to keep AI out of the building think they’re protecting their kids, when in reality, these kids are being left behind.

An example of what can happen when kids are taught to use AI responsibly comes in the form of a 14-year-old boy in Texas. To create greater access to diagnoses in his community, he built an AI app that screens for heart disease in seconds. That’s what’s possible when a child is taught to build with AI instead of just passively consuming it. 

There is no time to waste

Somewhere in a small town in Tennessee, a student is sneaking AI under the desk in a classroom where it’s ‘blocked’. They will use it unsupervised, they’ll try to hide it for fear of getting into trouble, and in the end, all it will have been used for is cutting corners in their assignments. 

In my view, we can make a change before students are put at a disadvantage. We need classrooms to be AI-native and provide access to tools that are safe, so students don’t default back to unregulated tools. For this, we need to redesign AI to emphasize education around critical thinking, complex building and ultimately, protecting human agency. The goal has never been for AI to think for a child, but for the child to think better.

Right now, our schools are not ready, and our teachers are not ready, and there is no plan in place to change that. But this generation doesn’t have the time to wait for them.

The opinions expressed in Fortune.com commentary pieces are solely the views of their authors and do not necessarily reflect the opinions and beliefs of Fortune.