Every few weeks, another student describes the same scene. A paper comes back not with a grade but with an accusation, because a detector has assigned the work a number, the number is high, and that number has become the entire case against the student. The accused is now asked to prove a negative, to show that the words on the page genuinely began with them. This is the wrong way to run a school, and districts should abandon the practice. AI-writing detectors may belong in a teacher's private curiosity, but they have no place in a disciplinary hearing, because a student's record must never depend on a proprietary probability score they cannot inspect or contest.
It is worth starting with the fairest version of the other side, because the pressure behind it is real. Teachers face a flood of AI-assisted work, and telling an honest draft from a generated one has become a daily guessing game. When cheating goes unpunished, the students who did the work honestly are the ones who lose. A detector, the argument goes, at least gives an overwhelmed teacher a place to begin a conversation, and any district that waves that concern away is not being honest about what its teachers carry.
But a detector fails in exactly the place where the stakes climb the highest. As one private signal that nudges a teacher to look closer, it may do little harm, but the moment it becomes evidence in a hearing, it imports a hidden error rate into a decision that can follow a student for years. That error, importantly, is not spread evenly across different kinds of students. When Stanford researchers ran seven widely used detectors against essays by non-native English speakers, the tools labeled more than half of those essays as AI-generated, and one of them flagged nearly ninety-eight percent. The same detectors correctly cleared more than ninety percent of the essays written by American eighth-graders. The machine is therefore not making mistakes at random. It is making mistakes that track who a student is, punishing those whose first language is not English for the plainer words they tend to choose.
It is also worth asking which companies have already walked away from this tool, because the answer says a lot. OpenAI, the company that built the model most students are accused of using, quietly shut down its own AI-text classifier in 2023, pointing to its low rate of accuracy. If the lab that trains these systems cannot reliably catch their output, a district has no business letting a vendor's score carry the weight of a formal charge.
Here, though, is the argument that ought to settle the question, and it has very little to do with accuracy. It is really a question of due process. When Vanderbilt University turned off Turnitin's AI detector, it explained that the feature had arrived with no insight into how it actually works. A student accused on the strength of a detector cannot examine the method, cannot question the training data, and cannot cross-examine the number being used against them. A charge the accused is simply unable to contest is not really evidence; it is an accusation wearing the costume of data. No fair proceeding, whether it unfolds in a courtroom or a classroom, should ever rest its verdict on a witness that no one is permitted to question.
The sheer scale of the practice is what makes the cost concrete rather than abstract. Vanderbilt actually did the arithmetic. Even the one percent false-positive rate the vendor advertised, applied across the roughly seventy-five thousand papers the university processed in a single year, would wrongly flag about seven hundred and fifty of them. Multiply that figure across an entire district, and then repeat it year after year, and the detector stops looking like a safeguard. It begins to look instead like a reliable machine for manufacturing false accusations.
None of this means that districts should simply shrug and accept whatever AI is doing to the classroom. It means that the proper response is a teaching response rather than a forensic one, which is a very different thing. Integrity is better rebuilt where it actually lives, in writing done under supervision in class, in required drafts and revision histories that reveal the work, and in asking students to defend their major assignments out loud. These methods ask a student to demonstrate their process, instead of asking an unaccountable program to guess at it from the outside. A district that cannot explain how its own evidence was produced has no business using that evidence to brand a child a cheat. So keep the detector out of the hearing, and return the responsibility to where it has always belonged, which is on teaching.