A pupil hands in an essay that is better than anything they have written before. The teacher wonders. The pupil, who wrote it themselves, notices the teacher wondering. Next time, they write something a little worse on purpose.
Young people interviewed for An AI Policy Isn't a Playbook, a report from the Center for Digital Thriving at Harvard's Project Zero, described exactly this: capping their effort, or performing mediocrity on purpose, so their work would not be flagged. Harvard Graduate School of Education highlighted the report on 6 October. The researchers call it “two-way suspicion”: teachers suspect pupils of using AI to cheat, and pupils fear being accused whether they used it or not.
What the survey found
The report draws on a national survey of more than 1,000 US public school teachers and principals, fielded in spring 2025, alongside interviews with educators and 31 young people.
A quarter of teachers and 29% of principals said AI had created “dilemmas or difficult decisions” in their schools and classrooms. Among those who described a dilemma, 74% of teachers and 69% of principals described situations where pupils were suspected of using AI to help with, or do, their work. Dilemmas were far more common higher up the school: 11% of elementary teachers reported one, against 29% in middle school and 44% in high school.
Only 18% of teachers and 24% of principals said their school had specific policies governing generative AI. That leaves teachers making up rules as they go, and pupils facing different expectations from one classroom to the next.
From referee to coach
The researchers describe a shift many teachers will recognise. Instead of coaching pupils through a new kind of learning, teachers find themselves refereeing, watching for AI use and deciding on the spot what counts as a foul. Both sides lose sight of the relationship that learning depends on.
The report does not argue that cheating is imaginary or that rules do not matter. Its point is that a policy cannot do the work on its own. Pupils need to understand where the line sits for a particular task and why, and teachers need ways to have that conversation before something goes wrong.
Five plays to try
The Center offers five free classroom routines, framed as plays a teacher can call:
- Find Your Feet: help pupils sort out what they value about their own work and where AI fits with that.
- Scout the Field: name the pressures that make pupils hide AI use.
- Align on the Line: map possible AI uses for a task and agree which are fine, which are not and which are unclear.
- Watch the Replay: look back at an AI dilemma and plan a better response next time.
- Name the Game: build shared words for new AI situations, so they can be discussed rather than suspected.
The third is the easiest place to start. Before the next extended piece of work, list five ways a pupil might use AI on it: checking spelling, generating ideas, explaining a concept, rewriting a paragraph, writing the whole thing. Ask the class where each one falls. Then agree the line together and write it on the task sheet.
Why this is AI literacy, not just integrity
A pupil who can explain why AI is fine for brainstorming in this task but not for drafting has understood something about learning, authorship and the purpose of the work. That is exactly the judgement AI literacy is meant to build.
Detection tools cannot give a pupil that understanding, and a false accusation can damage trust that took a year to build. A ten-minute conversation about the line, held before the work starts, protects both the learning and the relationship.
Sources: Harvard Graduate School of Education, The AI cheating crisis in schools isn't (only) what you think, 6 October 2026; Center for Digital Thriving, An AI Policy Isn't a Playbook. Survey of 1,018 US public school teachers and principals via RAND's American Educator Panels, April to May 2025.