A cheating problem both sides report, but neither side can prove
The Center for Digital Thriving's finding that 74% of teachers and 69% of principals volunteered an AI-related cheating issue when simply asked about their AI dilemmas, without being prompted specifically about cheating, suggests the concern is less a minority complaint than something close to a default experience among educators. That figure alone would be a familiar story about AI-enabled academic dishonesty, but the research's more striking contribution is what it found on the other side of that relationship.
Teachers in the study often cannot definitively prove a given piece of student work involved AI, and students have no reliable way to prove it did not. That symmetry of unprovable suspicion, rather than a clear-cut detection and enforcement problem, is the actual structural issue schools are navigating, and it is a meaningfully harder problem to solve than simply deploying better AI-detection tools would suggest.
What students told researchers about living inside that suspicion
The interviews with 31 teens ages 15 to 19, conducted between April and June 2026, surfaced a specific and counterintuitive behavior pattern: some students reported deliberately underperforming on assignments, producing visibly weaker work than they were capable of, specifically to avoid appearing too polished and therefore suspect. That is a direct, measurable cost of unprovable AI suspicion landing on students who were not actually using AI at all.
Students also described a second pressure pulling in the opposite direction, feeling pushed to use AI simply to keep pace with peers who were using it, regardless of whether they personally wanted to. Those two pressures together, fear of false accusation pushing some students to underperform and peer pressure pushing others toward AI use they might not otherwise choose, describe an environment where a student's actual AI use and their perceived AI use have become disconnected from each other in ways that distort normal academic behavior on both sides.
Lead researcher Beck Tench's framing of the dynamic
Beck Tench, the study's lead researcher, described the resulting pattern as a 'race to the bottom,' a framing that captures why this problem resists a simple technical fix. If the rational response to being suspected of AI use is to produce deliberately worse work, the incentive structure itself is pushing some students away from their actual best effort rather than toward it, the opposite of what any academic integrity policy is meant to achieve.
Tench's second observation, that 'the uses that are invisible to us aren't just the bad uses,' points at a related problem: legitimate, well-intentioned AI use by students is likely being flagged or suspected alongside genuinely dishonest use, because the detection and judgment tools available to teachers cannot reliably distinguish between the two. That inability to distinguish is arguably the core technical gap underlying the entire trust breakdown the research describes.
Why the research recommends coaching over policing
The study's recommendations are notable for steering away from stronger detection or enforcement and toward relational approaches instead: teaching AI literacy and ethics explicitly, having students discuss both productive and unethical AI uses as part of coursework, asking students open-mindedly how they use AI rather than assuming the worst, and including both teachers and students directly in developing school and district AI policy.
The framing Tench offers, positioning the student-teacher relationship as 'us-and-them' facing the broader challenge of AI in education, reflects a bet that rebuilding trust requires treating students as partners who help shape how AI gets used responsibly in their own education. Whether that relational approach actually reduces dishonest AI use as effectively as stronger detection tools would remains a genuinely open question. The research's clearer contribution is addressing the unprovable-suspicion problem that detection tools alone cannot solve.
What this means for district and institutional AI policy more broadly
For school and district leaders building AI policy, this research is a useful counterweight to the instinct to solve AI-related academic integrity problems primarily through better detection technology. If the core problem is a trust breakdown running in both directions rather than a pure detection gap, investment in better AI detectors addresses only half the actual dynamic, and may do little to address the deliberate-underperformance behavior the research specifically identified in student interviews.
The recommendation to involve both teachers and students directly in policy development is also a practical process point any district currently drafting or revising AI policy, including the two major districts that moved to restrict generative AI entirely this fall, should weigh seriously: policy built without the people who will actually live inside it day to day risks reproducing exactly the suspicion-without-proof dynamic this research describes, regardless of how the formal rules are written.
