It is 7:42 AM. A high school principal is staring at an AI-detector report flagging a junior's college essay as 87 percent likely to be AI-generated. The student is across the hall waiting to be called in. The parent is on email asking how the school plans to handle this. The detector vendor's own documentation says, in small print, that scores under 95 percent should not be the sole basis for an academic-dishonesty finding. The school's AI policy is two paragraphs long, was drafted in 2024, and does not mention detectors at all.

Bans and AI detection belong in the same conversation, and neither solves the problem. A ban sounds decisive, but students can still access AI elsewhere, teachers lose access to useful planning supports, and the school avoids the harder question: what kind of AI use helps or harms learning?

AI detectors create a different problem. OpenAI's current educator guidance says detector research did not show enough reliability for judgments with lasting consequences. It also warns that detector-style judgments can wrongly flag human writing and may disproportionately affect English learners or students whose writing is concise or formulaic.

The better direction is process evidence: draft history, conferences, prompt logs, AI-use notes, oral explanation, source checks, and assignment designs that ask students to make their thinking visible. The goal is not to catch students using tools. The goal is to know what the student understands.

NYC Public Schools and several other large districts have adopted a traffic-light model: red for prohibited uses, yellow for uses that require safeguards and professional judgment, and green for approved uses with reviewed tools. The value of this approach is specificity. It tells educators that AI is not one thing.

Red uses include cases where AI would replace legally or educationally responsible human decisions, such as special-education determinations, final grading judgments, or consequential pathway decisions. Yellow uses include student data analysis, diverse-learner scaffolds, sensitive translations, and student use. Green uses include brainstorming, organizing, drafting with review, summarizing non-sensitive information, accessibility support, and professional learning.

Treat the model as an adaptable thinking frame, not as a policy template to copy whole. Each district still needs local review, but the traffic-light structure helps teams ask the right first question: what exactly are we using AI to do?

Six challenge categories deserve operational responses rather than a generic risk register: reskilling teachers, bias and transparency, data privacy, technological barriers, curriculum integration, and dependence on technology.

Teacher reskilling means time for staff to practice with tools, compare outputs, identify failure modes, and redesign assignments. Bias and transparency mean asking what data a model was trained on, what it optimizes, who is likely to be misread, and how errors are appealed. Data privacy means approved tools, data minimization, and clear rules against entering student PII into unreviewed systems.

Pedagogical integration is the hardest part. AI should support the learning target rather than becoming the activity. If the tool removes the struggle that the assignment was designed to produce, the task needs redesign.

A 3x3 ethical matrix, drawn from MIT's AI + Ethics Curriculum for Middle School, gives educators a concrete way to slow down a decision. Choose three stakeholders and three values, then ask what each stakeholder gains, loses, risks, or needs under the proposed AI use.

Take an AI grading assistant as a worked example. The stakeholders are students, teachers, and families. The values are learning, fairness, and privacy. The matrix surfaces tensions that a yes/no tool decision hides — and it forces each stakeholder's perspective into the room before the decision is made.

A leadership team can use the matrix before adopting a tool, after a pilot, or when writing staff guidance. It should end in a decision, not just a conversation: proceed, proceed with safeguards, revise the use case, or do not proceed.

Policy becomes clearer under pressure, which is why AI dilemmas are a useful planning tool. The site's three flagship educator dilemmas pressure-test the most common AI policy edge cases. Each one is built to run as a twenty-minute staff discussion before a real version of the case lands in an administrator's inbox.

The leadership move is to use these scenarios before a crisis. A policy committee can read one case, identify what current policy would say, locate the ambiguity, and draft a rule that a teacher, student, substitute, parent, and administrator would interpret the same way.

AI Authorship Quandary

A policy stress test for student writing and disclosure.

From Ambiguity to Action

Turn values into rules educators can actually apply.

Educator Thought Experiments

Run policy dilemmas as staff discussion.