A Decision Framework for Suspected AI Misuse (opinion)

It is the last week of the semester and a faculty member is reading a paper that does not quite feel right. The citations are real, but two of them point to a textbook the class did not use. The middle of the paper reads in a different voice than the introduction. Running the paper through an AI detector, half out of curiosity, returns a high score. Now what?

The options are familiar to anyone who has been through this: Email the student and ask. Fail the assignment. Refer it to the dean of students. File an academic-integrity report. Drop a grade and say nothing. Each path carries different consequences for the student, procedural obligations for the institution and personal exposure for the faculty member. The syllabus does not say which path is which, and the institutional AI policy, if one exists, tends to say that AI use may be a violation and stops there. So, the faculty member improvises.

This is where most of the legal exposure in academia’s AI moment lives. The field has spent three years debating whether AI use constitutes cheating; the more pressing, largely unresolved question is what faculty and administrators are actually supposed to do when they suspect it. Most institutions have left that question unanswered. The result is faculty acting on instinct, students facing inconsistent processes and institutions absorbing risk that better procedure would prevent.

What follows is a decision framework, designed to be usable at a department meeting or Faculty Senate discussion by those without a law degree.

The Distinction on Which the Framework Rests

The framework starts with a distinction the Supreme Court has been drawing for more than 50 years. The 1975 decision Goss v. Lopez established that public school students have a constitutionally protected interest in their education that the state cannot take away without due process. Three years later, in Board of Curators v. Horowitz, the court extended that principle to higher education and drew the line between two kinds of decisions an institution may make about a student. Decisions about the quality of academic work, what Horowitz called “academic judgment,” receive substantial deference and require minimal process. Decisions that a student violated a rule, what Horowitz called “disciplinary determinations,” require more.

The reason matters. Grading decisions involve professional judgment about the work in front of the faculty member. Misconduct findings involve factual claims about what the student did and carry consequences that follow the student beyond the assignment: transcript notations, disciplinary records, suspension and expulsion.

Two recent cases show the principle from both sides. In Yang v. Neprash, a federal court declined to disturb the University of Minnesota’s expulsion of a doctoral student for unauthorized AI use on a qualifying exam, finding that the notice, hearing, advocate representation and appellate review the institution provided satisfied procedural due process. (The court dismissed without prejudice, also noting that the student had not exhausted state-law remedies.)

In Matter of Newby v. Adelphi University, a state court annulled an academic-integrity finding and ordered expungement because the same administrator who issued the violation also decided the appeal, rendering the appeal mechanism, in the court’s words, “inconsequential,” and because the university disregarded contradictory AI-detection results the student submitted. Different doctrinal routes (constitutional due process at a public institution, contractual fair procedure at a private one), same lesson: Courts look closely at whether the institution applied the right process for the consequence imposed.

That distinction is the foundation, and the framework below routes a faculty member’s suspicion through it, step by step.

Step 1: What Do You Actually Have?

Start with what the work itself shows, not with what a detector says. Detectors are contested at every level: whether they should be used at all, whether they reliably distinguish AI-generated from human writing and whether they produce false positives and false negatives. Some institutions prohibit their use entirely in misconduct proceedings, others permit them only as one input among many and almost none treat them as sufficient evidence for a finding. A faculty member relying primarily on a detector score is building a case on contested foundations, subject to cross-examination by a competent advocate on every one of the abovementioned grounds.

By contrast, the evidence that holds up is observable in the work itself. Hard indicators include citations to sources the student likely would not have used, such as a textbook other than the assigned one; content the student could not plausibly have known; fabricated references pointing to articles that do not exist; or a submission that contradicts work the student produced in class under direct observation.

Soft indicators include stylistic shifts midpaper, a tone that does not match prior work the student has produced in class and the specific tics of LLM-generated prose: em dashes used as punctuation throughout, frequent contrastive framing (“not X but Y,” “this is not just A, it is B”), a tendency to use bullet point–structured lists where prose would do, hedging openers (“it is important to note,” “it is worth considering”) and a polished, generic register that smooths over the rough edges most student writing carries.

Hard indicators are reasons to make findings; soft indicators are reasons to have a conversation. The faculty member who finds fabricated citations is in a fundamentally different position than one who has only a stylistic hunch.

Step 2: What Does Your Syllabus Say?

Whatever AI rule applies to this assignment should be in the syllabus. If it is not, the faculty member is on substantially weaker ground in calling the use a violation. Courts and institutional appeal bodies both look for whether the student was on notice of the rule they are accused of breaking, and vague language about “appropriate use” or “academic integrity” generally is not enough to put a reasonable student on notice that a specific AI use was prohibited. This is not about technicalities. It is about what fairness requires when consequences attach to a rule the student may not have known existed.

Step 3: Did the Student Fall Short of a Course Expectation or Violate an Institutional Policy?

This is the pivotal step, and it is the step most often skipped. A course expectation that a faculty member sets for an assignment differs from an institutional misconduct policy that the university establishes, and that difference determines everything that follows.

If the student fell short of an expectation the faculty member set for the assignment, such as “do not use generative AI to draft this essay,” the question is how the work measures against the criteria. The faculty member is exercising academic judgment about whether the submitted work meets the criteria on the rubric, and that is grading.

If the student violated an institutional academic-integrity policy, the question is whether the student committed misconduct. That is a different determination, made under a different framework, with different procedural requirements and different consequences. Faculty cannot turn a course expectation into a misconduct finding by relabeling it.

Many suspicions of AI misuse can be resolved at the course-expectation level without ever invoking misconduct procedures. The work did not meet the criteria, so the student gets a low grade or a zero, the same way they would for any other work that did not meet the criteria. If the student disagrees, the standard grade-appeal process is available.

Step 4: Who Sets the Process?

Once the faculty member knows which kind of question is on the table, the answer to who sets the process follows.

Faculty set the process for assigning consequences at the assignment level. The faculty member decides how to grade the work, what feedback to give and whether to use the situation as a teaching moment. The institution does not need to ratify a grade. The student’s recourse is the institution’s grade-appeal process.

The institution establishes the process for consequences for misconduct. Academic-integrity findings, transcript notations, course failure on integrity grounds, suspension and expulsion all run through the institution’s misconduct procedures, not through faculty discretion. Newby is partly a case about what happens when a single institutional actor wears two procedural hats and the appeal route becomes meaningless.

Step 5: What Process Does the Consequence Require?

Within whichever track applies, calibrate the process to the stakes. The higher the consequence, the more process the student is owed. A regrade on a single assignment can be handled informally. A zero on a major assignment warrants documented notice and an opportunity for the student to respond. A course failure on integrity grounds, a transcript notation or a suspension all require formal proceedings consistent with the institution’s misconduct policy and its constitutional or contractual obligations to its students.

Process is the institutional record that a court or appeal body will examine if the decision is later challenged. A decision made through a documented process, with the student given a meaningful chance to respond and an actual appeal route, is the decision that holds up.

What Institutions Need to Put in Place

The framework above works only if the institution provides the necessary infrastructure. Five categories of policy work matter:

1. Syllabus templates that specify AI rules at the course and assignment level, not just at the catalog level. The template should distinguish among assignments where AI use is prohibited, permitted with disclosure, permitted with limits (brainstorming only, drafting only, no generation of final prose), or required. Faculty should be able to select among options for each assignment rather than crafting language from scratch each semester, and the language should be drafted with the notice question from Step 2 in mind: Would a reasonable student understand what is and is not permitted?
2. Defined routing procedures that distinguish grading questions from misconduct questions, with clear entry points for each. Faculty should know where a Step 3 grading question lives (the rubric, the grade book, the grade-appeal route if challenged) and where a Step 3 misconduct question lives (an academic-integrity office or equivalent, with a documented intake process). Institutions that route everything through the same office, or that route nothing through any office and leave it entirely to faculty discretion, are the institutions whose cases will end up in court.
3. Training that helps faculty draw the Step 3 distinction in real time. The point of the training is not to make every faculty member a lawyer; it is to give faculty a clear question to ask themselves when suspicion arises: Do I have a grading question about whether this work meets the criteria, or do I have a misconduct question about what the student did? Most cases resolve at the grading level once faculty have language to use.
4. Explicit consequence tiers that map the process to the stakes at Step 5. An assignment regrade does not require formal proceedings. A course failure on integrity grounds does. Transcript notations, suspensions and expulsions require process consistent with the institution’s own policies and its constitutional or contractual obligations to its students. The tiers should be documented and trained on, not left to the administrator’s improvisation, case by case.
5. Structural separation between the person who issues a misconduct finding and the person who hears the appeal. This is the Newby lesson: An appeal that lands on the desk of the same person who made the original finding is not an appeal in any meaningful sense, and a court that sees that structure is likely to say so.

Underlying all of these is the standard of proof. Preponderance of the evidence is the most common standard for academic misconduct findings, but some institutions and some jurisdictions impose different standards, including clear-and-convincing evidence in particular contexts. The standard matters at the practical level because a detector score and a stylistic hunch may not clear a preponderance threshold, and they almost certainly do not meet a clear-and-convincing one. Faculty and administrators need to know what standard their institution requires before they make findings under it.

Back to the Faculty Member

The faculty member I described at the start of this piece has a clearer path than the syllabus suggested. The citations to the wrong textbook are hard indicators worth investigating, but the detector score alone is not. If an AI rule was in the syllabus and the work fell short of the expectations on the rubric, the faculty member can give a grade that reflects that, with documented feedback. If the faculty member believes the student violated an institutional integrity policy and a finding should attach to the student’s record, that goes through formal misconduct procedures, not through unilateral faculty action.

The broader debate about AI in higher education will continue. The narrower question of what to do when suspicion arises is one that institutions can answer now. Courts have been clear about what process should look like when consequences attach to a student’s record. Institutions that put that process in place will protect their students, their faculty and themselves. Those that do not will keep ending up in court.