Imagine your child comes home from school and tells you they used a new AI tool in class today. It helped them with their writing. It gave them feedback on their math reasoning. You might feel impressed. You might feel a little uneasy without being able to name why. You ask the question any reasonable parent would ask: what is the tool? Your child shrugs. Their teacher picked it. The district approved it. Somebody, somewhere, decided this was fine.
Now imagine asking a different question, one that almost nobody in the chain of decision making has a clear answer to: where does the data go?
Your child’s writing patterns, their error tendencies, the pace at which they learn, the moments they hesitate, the questions they ask when they think only the machine is listening. All of that is data. And in the current landscape of AI in education, the honest answer to where it goes is: almost nobody knows for certain.
The Infrastructure That Does Not Exist Yet
The infrastructure for AI data governance in schools does not exist in most districts. Not because administrators are negligent, but because the category itself barely existed three years ago. Schools were still building frameworks for managing student data in cloud-based platforms when generative AI arrived and changed the nature of the question entirely.
The old question was about storage. Where does student information live, who can access it, is it encrypted? The new question is fundamentally different. When an AI tool processes your child’s writing and reasoning patterns, does that data train the model? Does it improve the product for future users? Does it leave the school’s ecosystem entirely and enter a corporate dataset?
Most districts cannot answer this because most EdTech companies do not make the answer easy to find. Privacy policies run to dozens of pages. And the educators making purchasing decisions are not trained to evaluate these claims even when the claims are made clearly. The regulatory void is not a gap. It is a canyon, and children are standing at its edge.
The Classroom as Training Dataset
When your child struggles with a concept, that struggle has always been a private and developmental moment. A teacher might observe it, respond to it, adjust their instruction. But the struggle belonged to the child. It was part of their learning process, not a commodity.
AI tools that collect behavioral data during learning transform that relationship. Your child’s confusion becomes a data point. Their hesitation becomes a signal. The question they almost asked but deleted before submitting becomes, in some systems, a recorded event. Not because anyone set out to surveil children, but because that is how machine learning works. The model improves by ingesting patterns, and the richest source of patterns in education is the messy, deeply human process of a child figuring something out.
The Cyberpsychology of Being Watched Without Knowing
Surveillance research in psychology has established a consistent finding: when people know they are being observed, their behavior changes. They become more cautious, more conformist, less willing to take risks. This is true for adults. It is profoundly true for children in learning environments, where risk-taking and mistake-making are not incidental to learning. They are learning.
But the dynamic with AI data collection is more psychologically complex than traditional surveillance, because the child may not know they are being observed in this way. They experience the AI tool as a helper. They do not experience it as a system recording the texture of their cognition and sending it somewhere they cannot see.
This creates what cyberpsychologists call an asymmetric awareness environment. The AI system has full knowledge of what is being collected. The child has none, and no meaningful capacity to consent even if they were told. A ten-year-old cannot evaluate a data privacy policy. A fourteen-year-old cannot weigh the long-term implications of their learning patterns being incorporated into a commercial model. These are not failures of the child. They are features of childhood.
The Consent That Was Never Asked For
FERPA was written in 1974, designed to protect student education records that at the time meant transcripts, disciplinary files, and report cards. It was not designed for a world in which a child’s keystroke patterns and problem-solving sequences are captured in real time by a third-party AI system.
Parents are theoretically part of this process. In practice, most do not know which AI tools their child’s school has adopted, what data those tools collect, or where it goes. The consent forms, when they exist, are bundled into beginning-of-year paperwork alongside emergency contacts and field trip permissions. They are signed, not read.
Why This Matters
This is not an argument against AI in education. AI tools have genuine potential to personalize learning and identify struggling students before a human teacher could notice. The technology is not the problem. The problem is the unconsciousness with which it is being deployed. The speed that outpaces understanding. The assumption that because a tool is useful, its data practices are acceptable.
The data being collected is not abstract. It is a map of how your child thinks. How they approach problems. Where they get stuck. What they give up on. It is an intimate portrait of cognitive development, captured at scale, during the most formative years of a person’s life.
This matters because we are setting precedents right now about what is acceptable to collect from children in educational settings. The norms we establish in the next few years will determine the data landscape an entire generation grows up inside. And it matters because children deserve to learn in environments where their mistakes belong to them. Where their confusion is not a commodity. Where figuring something out is protected as a developmental experience, not harvested as a training signal.
You did not consent to this. Your child did not consent to this. And the system, as it currently operates, was not built to ask.
Leave a Reply