The teacher's claim is about a standard: a B-/C+ writer didn't produce graduate-level rhetoric over a weekend, and accepting the essay implicitly tells the rest of the class that the standard moved. The parent's claim is about fairness: the school never prohibited the tool, and punishing a student under a rule that didn't exist on Monday is exactly the kind of arbitrary authority parents are right to push back on. Alex's claim is about learning: the essay reflects work he genuinely did, even if the work looked like a two-hour back-and-forth with a model rather than four hours alone with a notebook.

None of these is a bad-faith move. The teacher is enforcing a developmental promise — that the grade tracks what the student can do. The parent is enforcing a rule-of-law promise — that consequences require advance notice. The student is enforcing a learning promise — that engagement matters more than the route by which engagement happened. They are all correct, and the system gave them no way to all be correct at the same time.

The teacher's instinct is to reach for the AI detector. That instinct is going to keep producing the wrong outcomes for as long as it stays the first move. The detectors do not work well enough to carry the weight people want them to carry.

The Stanford finding is the most important fact a teacher can know before flagging a single essay. Even in cases where AI involvement is real, a detector score is a probability, not a verdict, and the probability is calibrated against a population that doesn't match many real classrooms. Vanderbilt's Brightspace team disabled Turnitin's AI-detection feature in August 2023 for exactly this reason, citing both the false-positive rate and the absence of meaningful transparency about how scores are produced.

This doesn't mean detection has no role. It means the role is "this essay deserves a closer human read," not "this essay is AI-generated and the student lied about it." The downstream conversation should be a conversation about the work, not a tribunal built on a number.

Source: Liang et al., "GPT detectors are biased against non-native English writers" (Patterns, 2023)

Source: Vanderbilt Brightspace, "Guidance on AI detection and why we're disabling Turnitin's AI detector" (August 2023)

Schools talk about authorship as if it were a binary: you wrote it or you didn't. The history of writing has never worked that way. Politicians publish books they didn't write a sentence of. Academics publish papers heavily revised by reviewers, copy editors, and labmates. Pop songs list a dozen writers in the credits. The convention in each of these domains is not "the named author did every word" but "the named author is accountable for every word."

That distinction — production versus accountability — is the lever student-facing policy is missing. A student who runs a draft past a peer tutor, takes the tutor's structural suggestions, rewrites the introduction four times, and submits the result is the author of that essay in every meaningful sense, even though dozens of judgments were made by someone else. The tutor's contributions become part of the student's work the moment the student understands and endorses them. The same logic can apply to AI assistance, but only if there is a clear practice for endorsement.

The MLA and APA both updated their style guides in 2023 to treat generative AI output as a citable source — not as authorship, but as a contribution that has to be named. That move treats the problem the right way around: not "did the AI write part of this?" but "is the AI's contribution disclosed so a reader can evaluate the work honestly?"

If the school's bar is production, AI-assisted work is automatically suspect. If the bar is accountability, the test is whether Alex can sit across from his teacher and defend his thesis, anticipate counter-arguments, and explain why he chose this evidence rather than that. That test scales to any tool. It does not require a detector.

Source: MLA Style Center, "How do I cite generative AI in MLA style?" (2023)

Source: APA Style Blog, "How to cite ChatGPT" (2023)

The parent's strongest argument is procedural: no rule, no violation. They're not wrong. A school cannot enforce a standard it never published and never trained its staff to apply consistently. The teacher's discomfort is real, but a teacher's private standard is not a policy. Treating it as one — punishing a student for failing to anticipate the teacher's unwritten line — is the kind of move that erodes trust faster than any AI controversy does.

The harder version of this argument: the syllabus didn't say "no calculators in math class" either, but no one would defend a student who hid a graphing calculator on a no-calculator AP exam. The difference is that the calculator rule was clear at the test level, supported by long-standing professional norms, and uniformly applied. The AI situation has none of those scaffolds yet. A school that wants to defend a standard has to build the scaffold first.

State and district policy is finally catching up, but the catch-up is uneven. Tennessee became the first state in March 2024 to require all districts to publicly post their AI use policies. Ohio's House Bill 96, signed in August 2025, mandates formal AI policies for every public district by July 1, 2026. NYC Public Schools released the most detailed U.S. district framework in March 2026 after twenty-five rounds of feedback with more than a thousand stakeholders. None of that helps the administrator sitting in this meeting today, but it changes what counts as a reasonable excuse a year from now. The window for "we haven't gotten to it" is closing.

Source: Tennessee Department of Education, AI Guidance (2024)

Source: Ohio Legislature, House Bill 96 (August 2025)

Source: NYC Public Schools, Guidance on Artificial Intelligence (March 2026)

Most academic-integrity language was written when product was the only available evidence of process. If the essay was good, the assumption was that the work behind it was good. AI severs that assumption. A student can now produce a strong product after a thin or thick process, and the product itself doesn't tell you which one happened.

Honest assessment after that severing has to do something the old assessment never had to do: ask about the process directly. A short oral defense, a process journal, a "show your prompts" requirement, a brief in-class follow-up question — any of these recovers the signal that the essay alone no longer carries. None of them require an AI detector. All of them produce a record the school can defend if challenged.

The simplest and most defensible rule a school can hold is this: AI use is allowed unless an assignment explicitly forbids it, and any AI use must be disclosed. The disclosure can be a short paragraph at the end of the work — what tool, for what purpose, with which prompts. It doesn't require new technology, doesn't require detection, and doesn't require the teacher to play forensic investigator. It does require a culture in which disclosure is the unremarkable default, not the confession that gets you in trouble.

The schools getting this right are pairing the disclosure norm with two commitments. The first: students who disclose AI use honestly are evaluated on their work; they are not punished for the disclosure itself. The second: assignments where AI use would defeat the assignment's purpose are clearly marked as AI-restricted, with the reason explained ("this is a timed in-class essay because I am assessing your ability to think under pressure," not "no AI because cheating is bad"). That pairing changes the incentive structure. Concealment becomes the only path to trouble.

Gert Biesta's framework, which surfaces across this site (see AI Ethics for the broader treatment), identifies three purposes of education: qualification (knowledge and skills), socialization (induction into shared norms), and subjectification (becoming an autonomous moral agent). AI assistance affects each purpose differently, and a single AI policy that doesn't notice the difference will mishandle most of the cases it touches.

For qualification — discrete competencies, factual recall, technical skill — AI assistance is often genuinely useful and often genuinely indistinguishable from older forms of help. A student using AI to understand a math concept is doing what a good tutor would otherwise do. Restricting AI here trades a small integrity gain for a real learning loss.

For socialization — learning how to participate in a discipline, how a historian argues, how a scientist defends a claim — AI assistance is more ambiguous. The student who only ever reads AI summaries of original sources is not being inducted into the discipline; they are being inducted into a derivative of it. The remedy is not to ban the AI but to design assignments where the underlying engagement is visible.

For subjectification — becoming the kind of person who can think independently, hold a position under pressure, and revise it honestly — AI assistance is the most fraught. Coelho and colleagues, writing in the British Educational Research Journal in 2025, named the danger as a "placebo effect" (the appearance of independent thought without the actual development of it) paired with a "nocebo effect" (the student's retreat from the demanding work of becoming a thinker because the offload feels safe). The Authorship case sits closest to subjectification — what's at stake is Alex's development as an autonomous reasoner, not just whether the essay is good.

Source: Coelho et al., British Educational Research Journal (2025)

Source: Gert Biesta, The Beautiful Risk of Education (Routledge, 2014)

NYC's framework (covered in more depth on AI Ethics) sorts uses into three buckets: red (prohibited), yellow (conditional, requires active educator judgment and training), and green (approved, with normal data-privacy review). The framework is built around the idea that an AI rule has to be a rule about specific use cases, not a slogan about AI in general.

Alex's essay sits squarely in yellow. The assignment isn't AI-restricted on its face — it's a take-home literary analysis, which is a category where AI assistance is plausible and not automatically corrosive. The teacher's role under a yellow regime is exactly what the teacher in this case wants to do: read the work, talk to the student, and use her professional judgment about whether the engagement was real. The mistake the case reveals is not the teacher's instinct. It's that the school sent the teacher into the meeting without the policy that would have made her judgment defensible.

A workable AI policy doesn't need to be long. It needs to answer five questions in language a sixteen-year-old can read and a parent can quote back at you. Most of the trouble in cases like Alex's traces to a policy that punts on one of these.

A policy that answers these five questions in clear language defuses ninety percent of the Alex cases before they happen. Teachers know what to flag and how to handle it. Students know what they're being held to. Parents have something to read instead of something to argue about.

The single move most likely to make the Alex situation worse is also the one administrators are most tempted to make: apply a new rule to the case in front of them. The temptation is understandable. The case is the reason the rule is needed. Punishing this case under the rule that this case inspired feels like justice. It isn't.

Retroactive enforcement tells every other student in the building that the rules of the game are whatever the school decides they are after the fact. It guarantees a legal challenge and, more importantly, a culture in which families stop trusting the institution to behave predictably. The right move on the case is the move that the existing (absent) rule allows: resolve it through conversation, document the resolution, and use the case explicitly as the trigger for policy work whose results will apply to the next student, not this one.

The schools handling AI well are quietly making a pedagogical shift that the policy conversation hasn't quite caught up to. They are teaching disclosure as a writing skill — part of how a student learns to be honest about how their work came to be — rather than as a confession mechanism. The disclosure paragraph at the end of an essay becomes a place to think about process. Students start naming where they got stuck, what they asked the AI, what the AI suggested, what they kept, what they rejected.

This is closer to how professional knowledge work actually operates. A scientist writes a methods section. A journalist names their sources. A historian footnotes the archive. The disclosure paragraph is the student-writing analog of all those moves, adapted to the tool that's now in the room. Treating it as a skill rather than a punishment is the bet that the long-run interest is in honest writers, not catchable cheaters.

The policy point isn't that AI is bad for thinking. It's that schools have to design for the version of AI use that builds independent capability and against the version that quietly replaces it. Disclosure is the cheapest, most repeatable mechanism we have for telling those two versions apart.

Source: Microsoft Research, "The Impact of Generative AI on Critical Thinking" (2025)

My position on the Alex case, after walking through it from every chair:

None of this dissolves the underlying tension. Some students will use AI in ways that the school, on reflection, decides it doesn't want to permit. Some assignments will need to be redesigned because AI assistance does corrode them. Some parents will keep arguing that the school is overreaching. The work doesn't end. But the work changes character — from a series of one-off discipline cases handled inconsistently by exhausted teachers, to an institution that has decided what it stands for and built the scaffolding to hold the line.

The Authorship Quandary is not really about Alex. It is about whether the school is willing to do the slow institutional work that makes Alex's case rare instead of typical. The answer, in most schools, is currently no. The cost of that no is being paid in classrooms every day.

Matthew A. Zinn. "The AI Authorship Quandary." The Examined Classroom, February 14, 2024. (Original publication; this article is the expanded internal treatment.)

Weixin Liang, Mert Yuksekgonul, Yining Mao, Eric Wu, James Zou. "GPT detectors are biased against non-native English writers." Patterns 4(7), July 2023. DOI: 10.1016/j.patter.2023.100779.

Vanderbilt University Brightspace. "Guidance on AI Detection and Why We're Disabling Turnitin's AI Detector." August 16, 2023.

New York City Public Schools. "Guidance on Artificial Intelligence." Released March 24, 2026.

Tennessee Department of Education. "Department of Education Announces Guidance on Artificial Intelligence Use in TN Schools." March 13, 2024.

Ohio General Assembly. House Bill 96, signed August 2025. Mandates formal AI policies for every public district by July 1, 2026.

Higher Education Policy Institute. "Student Generative AI Survey 2025." Reports 92% of UK undergraduates using AI tools, up from 66% in 2024.

Digital Education Council. "Global AI Student Survey 2024." Reports 86% of students globally using AI in their studies.

Coelho et al. "Generative AI in schools: placebo and nocebo effects on subjectification." British Educational Research Journal, 2025.

Robert Sparrow and Gavin Flenady. "Bullshit Universities: The Future of Automated Education." AI & Society 40, 2025, 5285–5296.

Gert Biesta. The Beautiful Risk of Education. Routledge, 2014. Source of the qualification–socialization–subjectification framework.

MLA Style Center. "How do I cite generative AI in MLA style?" Modern Language Association, 2023.

APA Style Blog. "How to cite ChatGPT." American Psychological Association, 2023.

Microsoft Research. "The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence." Knowledge worker survey, 2025.

European Union. Regulation (EU) 2024/1689 (the AI Act). In force August 1, 2024. Classifies education as a high-risk domain under Annex III.