Most AI governance for schools does one thing: it tells you what already happened. Real governance shapes what happens next, inside the conversation, while a student is still typing.
It’s 9:40 on a Tuesday night. A ninth-grader has a paragraph due tomorrow and a document that’s been blank for an hour. She opens a chatbot and types: write my introduction about the causes of World War I.
Somewhere, a dashboard notes it. A weekly report will eventually show that a student used a generative AI tool during homework hours. An administrator might see it on Friday. Maybe it gets flagged.
But by then the paragraph is written, by the machine, and the student has learned the one lesson we didn’t want her to learn: that the fastest way through thinking is to skip it.
This is the gap at the center of almost every AI governance product for schools on the market today. Most of them govern nothing. They watch. And watching, however sophisticated, arrives too late to matter.
Monitoring answers a question no student is asking
Today’s market leaders approach student AI use the way an old content filter approached the web: from the network, after the fact, in aggregate. They can tell you that AI use is up this semester. They can surface which tools are popular. Some will even flag a suspicious spike before a due date.
All of that is visibility. It’s genuinely useful for a compliance report and genuinely useless in the moment that actually shapes a child’s learning, the moment between the question and the answer.
Visibility is retrospective by design. It describes a decision the student already made. Governance is different: it participates in the decision while it’s still open.
We think that distinction is the whole game. So we build for the moment, not the report.
Governing the conversation, not the network
When a student in a district running IMTLazarus opens an AI tool, they don’t drop into an open-ended answer machine. They enter an environment the district has shaped, one built around a simple conviction: AI should make students think more, not less.
Three ideas do most of that work.
The Socratic Tutor. Instead of handing over a finished paragraph, the AI is instructed to do what a good teacher does… ask the question back. What do you already know about the alliances in 1914? What’s your first sentence going to argue? The student still gets help. What they don’t get is the answer with the thinking removed. The tool becomes a tutor that refuses to do the homework, which is exactly what a tutor is for.
The Prompt Credit Economy. Unlimited, frictionless AI teaches students that thinking is cheap and infinite. It isn’t. Giving each student a considered budget of prompts changes the posture entirely: they slow down, they write a better question, they try the problem before they reach for help. Scarcity, used on purpose, restores deliberation.
Cognitive Security. Schools pour enormous effort into protecting students’ data and devices. Far fewer are protecting the thing AI most quietly erodes — a young person’s developing capacity to reason, struggle, and arrive somewhere on their own. Cognitive security treats that capacity as an asset worth defending. Not by banning AI, which is neither possible nor wise, but by making sure every interaction with it leaves the student’s own thinking stronger, not outsourced.
None of this shows up on a network dashboard, because none of it happens on the network. It happens in the conversation — the one place monitoring tools can’t reach.
“Visibility” is often just surveillance with better branding
There’s a quieter reason we resist the monitoring model, and it’s worth saying plainly. A tool whose core function is to record everything a child types is a surveillance tool, however friendly the interface. It changes the relationship between a school and its students, and rarely for the better.
Governing the interaction lets us do the opposite. We don’t need to log every keystroke to shape a healthier one. The goal isn’t to know more about students; it’s to build an environment where the easy thing and the good thing sit closer together. That’s a posture of trust and pedagogy not one of watching.
What AI governance for schools really requires
Adopting this view costs something. It’s harder to govern a conversation than to chart it after the fact, and it asks districts to make real pedagogical choices — what the tutor should refuse to do, how large a prompt budget should be, what “thinking more” looks like in a tenth-grade chemistry class versus a second-grade reading group.
But those are the right questions. They’re the ones a curriculum team should be answering anyway, and they’re far more useful than a Friday report confirming that yes, students are using AI. Of course they are. The question was never whether. It was always how, and whether anyone was shaping the how while it still mattered.
What’s next
This fall we’re bringing these ideas together in a single AI Governance release built for exactly this: shaping the live interaction, inside the tools students already use, with controls that sit with educators rather than on a distant dashboard. We’ll have more to share in September.
Until then, one question is worth carrying into back-to-school planning: when your district talks about “AI governance,” does the tool you’re considering change what happens in the conversation or does it just tell you about it afterward?
Because those aren’t two versions of the same thing. One is visibility. The other is governance. And students only benefit from one of them.
