The Rules Vacuum: AI Adoption Outran the Rulebook

Cover Memo

This report draws on CHIP50 Wave 38.6 (total unweighted N = 23,462; fielded June 21–July 13, 2026). It uses two distinct populations, kept separate throughout:

• Currently-enrolled students (school_recency = "currently enrolled", n = 2,214) for the items on instructors' stance, curricular integration, comfort asking questions, and how grades are structured.
• Teaching respondents (n = 496) for the single instructor-side item on whether AI has changed how they assess students. This is a small, self-identified group; its figures are reported only as toplines, with no demographic breakdowns (cells would suppress), and it is not the same population as the student sample.

All percentages and means are survey-weighted to national demographic targets; unweighted N is a reliability cue only. Figures are self-reported. Descriptive differences are labeled as such.

Key Takeaways

Introduction

Adoption ran ahead of rules. As the flagship in this series showed, roughly two-thirds of currently-enrolled students already use AI for their coursework — a level of uptake that arrived in barely two years. Formal guidance did not move nearly as fast. National data show the same lag: a 2026 Lumina Foundation–Gallup study found that 42% of college students say their school discourages AI and 11% prohibits it outright, while only about four in ten say students are encouraged to use it — a gap the researchers summarized as AI being "a routine part of college students' academic work, even as institutional guidance has not fully caught up."¹

CHIP50's Wave 38.6 supplement lets us see that gap from inside the classroom. It asks currently-enrolled students where their instructors stand, whether AI is part of their required courses, and how their grades are structured — and asks teaching respondents whether AI has changed how they assess. The picture is not one of confusion so much as of absence: students mostly know the rules in their individual classes, but those rules are a patchwork of personal instructor preferences rather than a coherent institutional policy, and they lean toward discouragement even as the assessments that determine grades remain wide open to the technology.

Discouraged, Not Guided

Where students can see an instructor's stance, it is far more often discouraging than encouraging. Only 11% of currently-enrolled students say their instructors mostly encourage AI use. Nearly half — 46% — report a mix, with some instructors encouraging it and others not, and 32% say their instructors mostly discourage it. Just 11% are unsure what their instructors want (Table 1).

That last number matters, because it reframes the problem. The issue is not that students are operating in the dark: only about one in ten can't read their instructors' expectations. The issue is that those expectations vary instructor by instructor and tilt against use — a student can be encouraged to use AI in one class and penalized for it in the next. For a technology two-thirds of students already rely on, the governing reality is discretion, not policy: a discouraging-to-mixed default set one classroom at a time. That default is recognizable across groups — no one is in the dark about it — but perceptions of it are not perfectly uniform: net of controls, Republicans and men perceive somewhat more encouragement than Democrats and women, and Black students perceive their instructors as more encouraging of AI than other students do, in keeping with this series' broader finding that Black students engage AI more readily (urbanicity and region make no difference — regional student AI adoption spans only 59.1% in the Northeast to 67.9% in the West, with no region significantly different from another). These are modest tilts on a shared perception rather than different realities — the discouraging-to-mixed patchwork is what students across the board describe, a structural feature of how institutions have responded more than an artifact of who the student is.

Comparative context. The tilt toward discouragement matches Gallup's national finding that a majority of students say their school discourages or prohibits AI, against a minority who feel encouraged.¹

Not in the Curriculum

AI is something students bring to their coursework, not something their courses are built around. Half of currently-enrolled students — 50% — say AI is not integrated into any of their required courses. Another 30% say it's in some required courses and just 10% in most; 10% aren't sure (Table 2). In other words, for the typical student, using AI is a private, self-directed practice layered on top of a curriculum that formally ignores it.

This is the structural counterpart to the discouragement finding. Institutions have neither embraced AI (by building it into required coursework) nor effectively walled it out; they have largely left it unaddressed at the level where it would matter most — the design of the courses and assignments themselves. The result is a two-track reality: an official curriculum that proceeds as if AI were not there, and an unofficial one in which most students quietly use it.

Two Sides of the Desk

The gap is not going unnoticed by instructors — but their response is individual, not institutional. Among teaching respondents in the survey, 51% say the availability of AI has already led them to change how they assess students — 20% "significantly" and 31% "somewhat" — while 49% say it has not (Table 3). Instructors, in other words, are adapting assessment on their own initiative, one course at a time, in the same decentralized way that produced the discouraging-to-mixed patchwork students describe.

What makes that adaptation urgent is where grades actually come from. Asked how much of their overall grade rests on each type of assessment, currently-enrolled students report that written papers and essays carry the most weight — more, on average, than in-person, closed-book exams, the one format AI cannot easily touch, and more than take-home exams (Table 4). The assessments that most determine students' grades are precisely the ones most exposed to AI. An institution that neither integrates AI into its courses nor updates the assignments those courses are graded on is, in effect, leaving the highest-stakes work in the most exposed format — and leaving individual instructors to resolve the contradiction assignment by assignment.

(A related access question — whether students are comfortable enough asking their instructors questions that they don't need to turn to AI instead — is taken up in the next report in this series; here it is enough to note that most students are comfortable approaching instructors, so the turn to AI is not primarily driven by fear of the professor.)

Conclusion

The story of AI in the classroom is not, at least yet, a story of clear rules being broken. It is a story of rules never having been written. Two-thirds of students use AI; one in ten is encouraged to; half sit in curricula that formally ignore it; and the grades that matter most still ride on the written work AI does best. Instructors see the problem and are changing their assessments — but individually, which is exactly the decentralized pattern that left students facing a different rule in every classroom in the first place.

None of this settles whether AI belongs in coursework; reasonable educators disagree, and this report takes no side. What it documents is a governance gap: a widely adopted technology met not with a considered institutional answer but with the accumulated improvisations of individual instructors. Closing that gap — whether by integrating AI, redesigning assessments, or setting genuine course-level policy — is a choice institutions have mostly not yet made. Until they do, the rules that govern AI in school will keep being written one syllabus at a time.

Appendix A — Methods

Data. CHIP50 Wave 38.6, fielded June 21–July 13, 2026; total unweighted N = 23,462.

Populations. Student items restricted to currently-enrolled students (school_recency = "currently enrolled", n = 2,214). The instructor item (ai_assess2) comes from a separate block of teaching respondents (n = 496); it is reported as a topline only, with no demographic breakdown, and is not drawn from the student sample.

Weighting. Survey-weighted to national demographic targets; unweighted N is a reliability cue only. Self-reported throughout.

Measures (verified coding). instructor_view — "Do your instructors encourage or discourage AI use?" (1 = mostly encourage, 2 = some do/some don't, 3 = mostly discourage, 4 = not sure). ai_course_integ — "Is AI integrated into any of your required courses?" (1 = yes, most; 2 = yes, some; 3 = no; 4 = not sure). instructor_comfort — comfort asking instructors questions (1 = very … 4 = not at all). ai_assess_matrix_1/3/4 — share of overall grade from in-person closed-book exams / take-home exams / written papers & essays (1 = none … 4 = a large amount). ai_assess2 — "Has the availability of AI tools led you to change how you assess students?" (1 = yes, significantly; 2 = yes, somewhat; 3 = no).

Appendix B — Data tables

Table 1. Do your instructors encourage or discourage AI? (currently-enrolled; weighted %)

Base: 2,214 currently-enrolled students. Weighted %.

Table 2. Is AI integrated into any of your required courses? (currently-enrolled; weighted %)

Base: 2,214 currently-enrolled students. Weighted %.

Table 3. Has AI led you to change how you assess students? (teaching respondents; weighted %)

Base: 496 teaching respondents (separate population; topline only). Weighted %.

Table 4. How much of the overall grade comes from each assessment type (currently-enrolled; weighted mean, 1 = none … 4 = a large amount)

Base: ~2,210 currently-enrolled students. Higher = larger share of grade. Weighted means. Written papers/essays — the most AI-exposed format — carry the most weight.

Demographic robustness (SES, urbanicity, region)

Race is integrated in the body ("Discouraged, Not Guided"): perceived instructor stance is uniform by race except that Black students perceive somewhat more encouragement (instructor_view −0.16, net of controls). On the personal-conduct measure (ai_banned, developed in Report 2), Black (+0.14) and Hispanic (+0.12) students report more cheating-adjacent use, and lower-education students more still. Urbanicity and Census region are null.

Appendix C — Question wording and response codes

Source: the Wave 38.6 Qualtrics instrument provided by the CHIP50 / COVID States team (the Wave 38.6 codebook, 2026-07-20), which is the authoritative record of item wording and stored numeric codes for this wave. Stems below are the codebook's condensed forms, not screen-verbatim text; response options and codes are verbatim. Where an item's stored code order differs from its on-screen display order, the stored codes are authoritative and are the ones used throughout this report.

• school_recency — "When did you last attend school or university?" (1 = Currently enrolled … 5 = More than 10 years ago). Code 1 defines the student base.
• instructor_view — "Do your instructors encourage or discourage AI?" (1 = Mostly encourage; 2 = Mixed; 3 = Mostly discourage; 4 = Not sure).
• ai_course_integ — "Is AI integrated into your required courses?" (1 = Yes, most; 2 = Yes, some; 3 = No; 4 = Not sure).
• ai_assess_matrix_1–6 — "What share of the overall grade comes from each assessment type?" (1 = None; 2 = A small amount; 3 = A moderate amount; 4 = A large amount). Items: 1 in-person closed-book · 2 in-person open-book · 3 take-home · 4 papers/essays · 5 projects · 6 low-stakes quizzes. Higher mean = larger share of the grade.
• ai_course — "Have you ever used AI for schoolwork?" (1 = Yes; 2 = No; 3 = Not sure).
• ai_assess2 (teacher block — a separate population, not the student sample) — "Has AI led you to change how you assess students?" (1 = Yes, significantly; 2 = Yes, somewhat; 3 = No).