Education trust is the willingness of students, parents, and teachers to depend on one another, and decades of school research break it into named parts such as competence, honesty, and respect. A 2026 Harvard report on generative AI in U.S. schools describes how that trust can erode: when teachers suspect students of using AI and students fear being accused whether or not they did, the report calls the result "two-way suspicion." The report is built on a survey of 1,018 teachers and principals fielded in April and May 2025, plus interviews with 31 teens in 2026, so it shows patterns and does not prove cause.
Key takeaways
- Trust research in schools uses two well-known models: five facets (benevolence, openness, reliability, competence, honesty) and four (respect, competence, personal regard, integrity).
- In the Harvard survey, 25 percent of teachers and 29 percent of principals reported a dilemma or difficult decision about generative AI, and cheating dominated the stories.
- Only 18 percent of teachers and 24 percent of principals said their school had specific generative AI policies.
- Students in the interviews described fear of false flags from AI detectors, and some described teachers' own AI use as hypocrisy.
- The evidence is a national survey of adults plus a small interview set, so it describes experiences, not the effect of any single tool.
What researchers mean by trust in schools
In education research, trust is usually defined as a willingness to be vulnerable to another party based on confidence that the party will act in one's interest. A 2021 scoping review in Frontiers in Psychology examined 79 peer-reviewed K-12 studies published between 2000 and 2020. Of the 63 that gave an explicit definition, 55 treated trust as willingness to take a risk on the other party.
Two models dominate. Megan Tschannen-Moran and Wayne Hoy's 2000 review in the Review of Educational Research describes five facets: benevolence, openness, reliability, competence, and honesty. Anthony Bryk and Barbara Schneider's work on Chicago schools describes relational trust through four: respect, competence, personal regard, and integrity. The scoping review notes a measurement problem. It found 32 scales, analyzed 26, and concluded that most studies measure how trustworthy people appear to be rather than how much risk someone actually takes.
What the 2026 Harvard report measured
The Center for Digital Thriving at the Harvard Graduate School of Education published "An AI Policy Isn't a Playbook" in 2026. Its survey drew on RAND's American Educator Panels: 719 teachers and 299 principals, weighted to represent public school educators nationally. Teacher invitations went out on April 28, 2025, and principal invitations on April 21, 2025. Completion rates were 51.7 percent for teachers and 30.2 percent for principals, and the authors caution that teachers and principals are not matched pairs from the same schools.
Asked about dilemmas or difficult decisions involving generative AI, 25 percent of teachers and 29 percent of principals said yes. Among teachers the share rose with grade level: 11 percent in elementary school, 29 percent in middle school, and 44 percent in high school. Of the situations described, 74 percent of teacher accounts (171 total) and 69 percent of principal accounts (85 total) were coded as cheating-related.
Two-way suspicion, in the report's words
The report defines two-way suspicion as two parties both suspecting each other of using AI, eroding trust regardless of whether AI is actually used. The student quotes show the mechanism. One 15-year-old said students avoid putting in full effort because a polished piece of work risks being flagged. An 18-year-old described running a group paper through four AI detectors, and one flagged writing the student knew was typed by hand.
A second strain concerns fairness between the two sides. Educators and students both described students noticing teachers using AI to build lessons or grade work while asking for original student writing. The report says students can read this as hypocrisy and that it can erode trust. Only 18 percent of teachers and 24 percent of principals reported specific generative AI policies, which leaves many teachers to improvise.
Where AI presses on each facet of trust
The mapping below is an illustrative reading, not a finding of the Harvard report. It pairs the five Tschannen-Moran and Hoy facets with situations the report or standard school practice raise, to show where a school might look.
| Facet | Illustrative pressure point | Question a school can ask |
|---|---|---|
| Benevolence | A student flagged by a detector is treated as a suspect before being heard | Is there a process that assumes good faith until evidence says otherwise? |
| Openness | Teachers use AI for lessons but students are not told | Do teachers explain their own AI use as well as students' limits? |
| Reliability | Rules differ between classrooms, so the same act is allowed in one room and penalized in another | Are expectations consistent across a grade or department? |
| Competence | Educators asked to judge AI use without training or school policy | Do staff have guidance on what evidence is reasonable? |
| Honesty | Students deny use when confronted, or teachers imply certainty a detector cannot give | Are accusations tied to evidence beyond a single score? |
What the evidence does not show
The survey covers U.S. public school teachers and principals only, and the student material is 31 interviews recruited through a youth organization and school networks, chosen for variety rather than to represent all teens. No part of the report measures whether trust in a particular school rose or fell after an AI policy was adopted. The teacher and principal survey also dates from spring 2025, so practices may have changed since. For Canadian readers, school boards and provinces set their own policies, and the survey says nothing about them.
This article reports research and is not legal advice. Disciplinary procedures, student records rights, and AI rules differ by district, state, province, and school board, and a school's own policy or a qualified adviser governs a specific case.
FAQ
Is education trust the same as an education trust fund?
No. In this article it means trust between people in schools: students, teachers, parents, and leaders. Savings vehicles and legal trusts are separate topics governed by tax and trust law.
Do AI detectors settle whether a student cheated?
In the Harvard interviews, a student described a detector flagging writing that the student knew was original. A detector result is a signal at best, and fair process usually involves a conversation and other evidence.
Is two-way suspicion proven to lower grades or learning?
No. The report describes the pattern through survey answers and interviews. It does not measure learning outcomes or test what happens after a school changes its approach.
Why does the scoping review say trust is hard to measure?
Many scales capture how trustworthy people seem, such as caring or competent, instead of how much risk someone takes in relying on them. The authors of the review recommend measuring the vulnerability itself.
Related reading
- What AI is used for in classrooms
- State education codes and AI
- Impact and potential of AI in education
- OpenAI's education products and age rules
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Tench B, Weinstein E, James C, with Starks A and Konrath S. An AI Policy Isn't a Playbook: Educators Need Agency, Not Just New Rules, Center for Digital Thriving, Harvard Graduate School of Education, 2026.
- Shayo HJ, Rao C, Kakupa P. Conceptualization and Measurement of Trust in Home-School Contexts: A Scoping Review, Frontiers in Psychology, 2021.
- Tschannen-Moran M, Hoy WK. A Multidisciplinary Analysis of the Nature, Meaning, and Measurement of Trust, Review of Educational Research, volume 70, issue 4, 2000
- Bryk AS, Schneider B. Trust in Schools: A Core Resource for Improvement, Russell Sage Foundation, 2002