What Wake Would Not Detect

On June 17, in a nondescript work session, the board of the Wake County Public School System sat down with a revised draft of the AI policy it plans to bring to a final vote in the fall.¹ Wake is the fifteenth largest school district in the country. Roughly one hundred sixty thousand students. Ten thousand teachers.¹ Whatever it decides about generative AI will, by simple weight, become one of the templates other districts read on their way to their own decisions.

The draft is unusual in the way that quiet documents sometimes are. It requires students to disclose, in writing, when they have used AI on an assignment. It calls undisclosed use academic misuse and points it toward the honor code. And, notably, it discourages teachers from running student work through AI detection tools at all.¹ Wake will not, as an institution, try to catch its students. It will ask them to tell it.

The district is not naive about what it is choosing to give up. Its own memo says the reason is that the detectors are technically unreliable, that their percentage scores are opaque, and that they misfire against students whose first language is not English.² That last one is the sentence that has quietly ended the debate in the offices of superintendents who watch the case law closely. The tools bias against the children who are already the most vulnerable in the room.

The case that appears in every version of Wake's own reporting is Eleanor Canina's. A freshman at Green Hope High School was accused, on the strength of an AI detector, of using generative AI on an English assignment. Her grade went to zero. She appealed. A second teacher looked at her actual work. The zero became a hundred.³ She then organized a petition and spoke to the school board. Wake's decision to move away from detectors is downstream of that afternoon.

The Detector Was Never The Right Instrument

Two thousand miles west, in Palo Alto, a district is spending a fraction of its year defending against a one hundred fifty million dollar federal civil rights lawsuit over the same thing. A tenth grader wrote an essay on The Crucible. Turnitin flagged seventy-six percent of it as AI generated or influenced. His father submitted more than eleven hundred pages of drafts, revision histories, and timestamps to prove otherwise. The district would not budge. The final grade dropped to a C. The family sued.⁴

The details vary. The pattern does not. A recent Fortune piece put the story in a single line: eighty-four percent of students use AI for homework, and only three in ten schools have rules for it.⁵ Detectors were built to close the gap between those two numbers. They were built to answer a question the tools themselves cannot answer, and to answer it in a single decimal number pressed onto a piece of work whose actual complexity is invisible to them. A percentage is not a story. It is a claim. And when a claim about a child's honesty rests on a black box that gets things wrong roughly as often as it gets them right, the courts and the school boards are, at long last, taking that seriously.

What Disclosure Can And Cannot Do

Wake's answer is the softer, harder one. Do not try to detect. Ask the student to say. It is the right move. It is also the beginning of the work, not the end of it.

Disclosure is a good norm because it puts the student in the position of accounting for her own process, in her own voice, before the adult has to guess. It is a good norm because it teaches, at a low cost, that the process is a thing worth talking about. It is not, though, the same as seeing what actually happened. A ninth grader can write on the top of her essay that she used ChatGPT for outlining and have used it for the sentences too. A senior can write that he used no AI at all and have used AI to write the disclosure. The instrument the disclosure is meant to be, in the end, is trust.

The Substrate Underneath The Policy

The layer we work on at Koan is the one that lives underneath both the detector and the disclosure. When a student writes, she pauses. She deletes. She reaches for a word and reaches again. She copies a sentence in from another window and, sometimes, she rewrites it in her own hand. None of that is visible to a detector after the fact. None of it is fully captured by a self-report. All of it is exactly what a teacher, sitting at her elbow with all the time in the world, would notice.

A good district cannot put a teacher at every elbow. What a good district can do is preserve the small, live evidence of process, so that when the disclosure and the finished work do not agree, the teacher has something more honest than a number to look at. Not to prosecute. To understand.

Fall 2026

When Wake takes its final vote later this year, the district will have done a rare and honest thing. It will have refused a bad instrument publicly. It will have named disclosure as its preferred alternative. The next question, and the one that will show up in the second year of the policy, is whether Wake can help its teachers actually see what its students are doing. Not through detection. Through the plain, slow work of watching learning happen.

If a district decides not to detect, what is it prepared to see?