Your District Has an AI Policy. Your Teachers Still Do Not Know What Is Allowed.

Walk into almost any school this fall and you will find the same arrangement. The district has an AI policy, or at least a statement. The students are using AI every day. And the teachers, asked what is actually permitted in their own classrooms, give you different answers depending on which hallway you are standing in. In one national survey, 60 percent of educators said their district's AI policy was unclear. The policies exist. The practice does not match them, and in a growing number of states, closing that gap has become the local board's job.

By early 2026, education departments in 31 states had issued AI guidance for K-12 schools. Ohio became the first state to require every district to adopt a formal AI use policy, with a deadline of July 1, 2026. On August 20, the U.S. Department of Education added guidance of its own, urging schools to evaluate classroom technology by the learning problem it solves, for whom and for how long it should be used, and what evidence shows that it works, with emphasis on educator judgment, transparency with families, accessibility, and the protection of student data.

Missouri's approach is the instructive one for boards. The Department of Elementary and Secondary Education has published a handbook of best practices, and the state is making Google's AI tools and training available to schools at no cost. But Missouri is a local-control state. DESE has been explicit that its handbook is guidance rather than policy, and that adoption decisions belong to each district and charter school. In other words, the state has handed local boards free tools, a reference document, and the entire responsibility.

Heading into this school year, many districts took a cautious path: adopt broad guardrails, encourage AI literacy, and leave the classroom-level decisions to individual teachers. That feels prudent. In practice it produces exactly the inconsistency teachers are describing, because a policy that delegates everything has decided nothing.

This is a pattern we see in every sector, and we have written about it before: a policy adopted without an implementation layer is not implemented. There is no rollout sequence, so teachers learn about the policy from a PDF. There is no training, so each teacher improvises. There is no measurement, so nobody knows whether practice matches policy until a parent complaint or a contested grade forces the question. AI did not create this failure mode. It exposed it at unusual speed.

A usable AI policy is short, specific, and answerable by a teacher in under a minute. It resolves four questions.

• Which tools are approved, and who approves new ones? An approved list with a fast review path. Without one, staff adopt tools faster than the district can vet them.
• What student use is permitted, by type of work? A simple tiered scheme (not permitted, permitted with disclosure, encouraged) that teachers set per assignment and students can see.
• What data may go into a tool? Student records, IEP details, and identifiable student work do not belong in a consumer chatbot. Staff need that stated plainly, with examples.
• How are families told? Which tools their children use, for what purpose, and how to raise a concern.

The hardest cases arrive as discipline. A teacher suspects AI wrote an essay, runs it through a detector, and the student is referred for cheating. Detection tools produce false positives, and research has found they misfire more often on writing by non-native English speakers. A district that disciplines students on a detector score alone is building a record it cannot defend.

The policy should say what evidence supports an integrity finding (drafts, version history, a conversation with the student about the work), who decides, and how a student contests it. Those are the same due-process fundamentals that govern any disciplinary matter. They apply here with extra force, because the technology is new, the evidence is probabilistic, and the consequences land on a transcript.

The board does not write the classroom rules. It owns whether the rules exist, are understood, and are working. Five questions will tell you where you stand. Can every teacher state what is permitted in their classroom, and would two teachers in the same building give the same answer? Has every staff member been trained, not just notified? Which AI tools hold student data right now, and under what agreements? How many academic integrity cases this semester involved AI, and how were they resolved? And when does this policy come back to the board for review, given that the technology will not wait for an annual cycle?

Most districts do not need a new policy. They need the implementation layer the current one never had: a rollout, training, a short list of classroom-ready rules, and a review date. That is a manageable fall project, and it is far cheaper than the alternative, which is discovering the gaps one contested grade at a time. Boards in other sectors face a parallel question about what to oversee and what to delegate, which we cover in AI Governance for Nonprofit and Healthcare Boards.