This dissertation examines the variation in generative AI (genAI) governance that students encounter at course level during rapid policy emergence. Campus guidance is increasingly available, but is distributed across AI tool pages, cybersecurity rules, academic-integrity statements, teaching-center resources, and syllabus templates, while pedagogical use largely remains subject to faculty judgment. Faculty therefore serve as a practical bridge between institutional intent and the rules governing ordinary coursework. Using an integrated Becher–Weick–Lipsky framework, the study examines how disciplinary cultures, interpretation under ambiguity, and front-line discretion are associated with classroom AI decisions. The design is quantitative and proceeds in two phases. Phase 1, a completed documentary analysis of 300 upper-division undergraduate syllabi from twelve disciplines at five Texas public R1 institutions, found visible AI rules in 163 syllabi, with rules appearing more often in soft fields than hard fields but showing little difference between pure and applied fields. Phase 2, a planned faculty survey at one of those institutions, will examine how reported sensemaking orientations, perceived guidance conditions, and discretionary classroom AI practices vary across disciplinary contexts, and whether reported orientations are associated with reported discretionary practice after accounting for disciplinary context and professional characteristics. Together the phases describe visible course-level AI governance and faculty-reported discretionary practice, and test which reported orientations are associated with that practice, indicating where institutions can establish clearer student-facing boundaries without eliminating appropriate faculty discretion.

Keywords: generative AI, faculty discretion, sensemaking, disciplinary culture, street-level bureaucracy, syllabus policy, policy emergence

Generative artificial intelligence is already in the classroom. Students are using it, and graduates are entering professional environments where AI fluency is increasingly expected.

Institutions face a genuine governance dilemma. A blanket permission rule would ignore disciplinary differences in knowledge, authorship, assessment, and professional preparation. On the other hand, wide-ranging prohibition would misrepresent a technology whose implications are still being realized. Broad or principled guidance, and in some cases deliberate silence, may be the most practical institutional response available while the area being governed is still developing.

That response transfers part of the governance burden to faculty. Faculty occupy a unique role in ordinary course decisions, including assignment design, permitted tools, disclosure rules, grading judgment, and conversations with students about expectations. Individual instructors exercise professional judgment within their disciplinary communities as they address areas where institutional guidance remains broad or unresolved. Analysis of the top fifty United States universities finds that AI policies often develop reactively, without a systematic framework, while instructor discretion remains central to course-level implementation (Alba, McIlwain, & An, 2026).

State policy provides a parallel point of comparison. The Texas Responsible Artificial Intelligence Governance, enacted in 2025 and effective January 1, 2026, established a statewide AI regulatory framework but excluded institutions of higher education from its definition of “governmental entity” (Texas Responsible Artificial Intelligence Governance Act, 2025). The exclusion does not prohibit universities from establishing their own policies, but it illustrates a broader governance pattern. Even as the state regulates AI, it leaves ordinary decisions about student use in university classrooms to institutional and ultimately faculty-level judgment. Institutions reflect this layered arrangement. Public guidance identifies approved or institutionally supported AI platforms while treating student use of generative AI as a course-level decision and encouraging faculty to communicate whether, when, and how AI may be used.

The stakes of this ambiguity fall most directly on students. At all five institutions in this study’s design, unauthorized AI use is classified as academic dishonesty carrying sanctions that range from a grade penalty through probation, suspension, and expulsion (Texas A&M University, 2025; Texas Tech University, 2024; The University of Texas at Austin, 2025; University of Houston, 2015; University of North Texas, 2025). The institutions vary, however, in how explicitly each has declared that AI misuse falls under the existing conduct framework. The same student act can be permitted practice in one classroom and a disciplinable offense in another.

These conditions are visible in the syllabi faculty publish. Statements collected at the study institution in a single semester include a public affairs course permitting AI but not to the point of “minimal thinking,” a two-sentence prohibition within the same college, a nursing rule grounded in privacy and professional standards, and a writing course prohibiting AI because assignments “develop the writer”. Other syllabi say nothing about AI at all, and the silences appear across several disciplinary clusters instead of one field alone. The public record shows no single course-level template resolving classroom AI practice across colleges. Faculty discretion works to address this one course at a time with documents showing different rules, but they reveal little about the faculty-reported orientations and contextual conditions associated with those student-facing decisions. This study examines which reported orientations, contextual conditions, and professional characteristics are associated with the classroom AI decisions faculty describe.

Approximately four years into widespread access to large language models, faculty responses remain varied, institutional frameworks remain broad or flexible, and the relationship between reported faculty orientations, disciplinary context, and the classroom AI decisions students meet remains underexplored. Faculty are not simply passive recipients of policy. They exercise considerable classroom discretion over how AI is understood, permitted, and practiced in their courses. Institutions have responded with broad governance frameworks, a rational strategy that preserves flexibility while placing significant interpretive burden on faculty. The literature establishes that faculty responses to generative AI are varied, consequential, and often patterned by discipline with numerous existing studies documenting those differences at scale, though institutional contexts vary (Chirikov, 2026; Buele & Llerena-Aguirre, 2025; Yang et al., 2026). What remains less explained is whether institutional ambiguity, disciplinary formation, and reported faculty interpretation are associated with the classroom AI conditions students encounter.

This study examines two connected lines of inquiry in sequence. The first is what student-facing generative AI governance is visible across institutions and disciplines. The second is how faculty-reported sensemaking orientations, perceived guidance conditions, and discretionary classroom AI practices vary across disciplinary contexts, and whether those orientations are associated with reported discretionary practice after accounting for context and professional characteristics. This first question carries its own sampling design, coding framework, reliability evidence, and inferential tests, and it stands as a substantive foundation rather than as background for the survey that follows.

The next step in this study is to disseminate a survey to instructors at one of the five institutions. The survey's closed items examine reported orientations, perceived guidance, and discretionary classroom practices. The instrument collects closed-item responses only. The process by which faculty arrive at a particular decision is not observed here and is the subject of a planned follow-up qualitative study.

The two phases will be reported sequentially and compared where they address overlapping institutional and disciplinary conditions. The study does not assume that the documentary and survey evidence represent the same individual faculty members or directly verify one another.

The operative research questions are:

The question of whether disciplinary context moderates that association is examined in Chapter 3 as a pre-specified exploratory analysis. It is not posed as a research question, because the study is not powered to estimate moderation at the sample size it anticipates. The analysis, the detection threshold it depends on, and the cautions that apply to its interpretation are specified there.

The theoretical framework identifies disciplinary context, faculty sensemaking, and discretion as analytically distinct but theoretically related dimensions of classroom AI governance. Becher and Trowler's (2001) academic tribes and territories framework provides the basis for examining whether classroom AI decisions differ across disciplinary cultures. Disciplinary membership is treated as an explanatory and grouping condition and not as a stage in a sequence, which allows the analysis to ask whether fields differ without assuming that fields cause the difference. Weick's (1995) sensemaking theory informs the measurement of how faculty interpret ambiguity and construct workable understandings of institutional guidance. In this design, sensemaking supplies measurable orientations and not an observed process. Faculty encounter guidance that is distributed, incomplete, and unevenly applicable, and the study records what they report about how they read it. Lipsky’s (2010) street-level bureaucracy theory provides the basis for examining discretionary classroom practice as an organizational consequence of local decision authority. Faculty do more than execute institutional guidance. Through syllabus rules, assignment boundaries, disclosure expectations, and grading decisions, they determine the operative conditions of a student’s legitimate AI usage. Taken together the three frameworks support an examination of whether disciplinary context and reported sensemaking orientations are associated with variation in faculty classroom AI practice.

AI-related matters remain unsettled across higher education. Faculty responses have had time to develop, even as institutional practices and support structures continue to evolve. That response is occurring in a context of widespread student AI use (Rong & Chun, 2024) and substantial faculty concern about its implications. Recent survey data indicate that 83 percent of faculty are concerned about students’ ability to evaluate AI critically, while 82 percent worry that students may become too reliant on it (Digital Education Council, 2025). Yet reducing faculty responses to resistance or enthusiasm may be overly simplistic. Familiarity with AI is increasing, but classroom integration has not kept pace. Sixty-one percent of faculty respondents had used AI, but among those who used it in teaching, 88 percent characterized that use as minimal to moderate (Digital Education Council, 2025). Sector surveys similarly identify gaps among institutional strategy, faculty support, and classroom practice (Robert & McCormack, 2025). This uneven engagement may therefore reflect the interpretive distance between evolving institutional expectations and local practice rather than simple compliance failure. Broad institutional AI policies still leave room for interpretation, and classroom governance varies with the professional discretion that disciplinary epistemological cultures help shape. That pattern predates AI, but the rapid emergence of generative AI makes it unusually visible. The claim examined here is therefore bounded to what faculty report and to what public course documents show.

This study works within clear boundaries. The documentary analysis is complete and draws on public syllabi from five Texas public R1 universities within a two-year window covering Fall 2025 and Spring 2026. The faculty survey will be conducted at one of these public institutions and will include faculty with instructional or course-design responsibility, including adjunct instructors. This study does not evaluate pedagogical soundness, student outcomes, actual student AI use, or causal relationships, and it does not reconstruct cognitive sensemaking as it occurs. Documentary claims are drawn from the collected public syllabi only, and survey findings will be bounded to the study institution. The documentary analysis carries its own research question rather than standing as background for the survey.

Artificial intelligence refers to large language model and generative systems such as ChatGPT, Claude, Gemini, and Copilot. The term does not include ordinary search engines, grammar checkers, plagiarism detection software, or older forms of algorithmic assistance. The disciplinary quadrant refers to the four-part Becher and Trowler classification, which crosses hard and soft knowledge with pure and applied orientations and is assigned through the survey distribution structure and the disciplinary classification used in the documentary analysis. Sensemaking is the process by which people construct a workable interpretation of an ambiguous situation so they can act (Weick, 1995). In this study, sensemaking will be examined through retrospective reports and related indicators rather than treated as directly observable cognition. It also does not imply movement toward AI adoption, since integration, prohibition, limitation, irrelevance, and continuity may each reflect how faculty interpret their circumstances. Discretionary practice refers to the classroom decisions faculty make when institutional guidance remains broad, incomplete, or still developing. Policy emergence describes the broader process through which those unsettled conditions become increasingly specified through institutional rules and local practice.

Chapter 2 explains why faculty interpretation and discretionary classroom practice are appropriate objects of inquiry. The review examines four dimensions of AI governance in the literature: the governance conditions faculty encounter, the disciplinary cultures that give those conditions meaning, the reported sensemaking orientations faculty bring to them, and the discretionary classroom practice that varies with both.

Higher education AI policies have tended to emphasize data privacy, security, academic integrity, and broad principles, while leaving many pedagogical decisions to departments and individual faculty. That pattern appears in syllabus evidence, faculty survey research, and sector-level governance analyses (Chirikov, 2026; Robert & McCormack, 2025; Jiang et al., 2025; Digital Education Council, 2025). Longitudinal evidence shows that movement away from blanket prohibition can coexist with continued assignment-level restriction. At one institution, university-required syllabus language shifted from 57% course-wide prohibition in Fall 2023 to 23% in Fall 2025, while most collected syllabi still prohibited AI for at least some assignments (Sabourin, 2026). Across more than 31,000 syllabi at one large public university, explicit AI regulation reached 55% of courses by Fall 2025, fully restrictive policies declined about five percentage points per year among courses with AI policies, and framing shifted from integrity toward learning effects and attribution (Chirikov, 2026).

This governance condition remains uneven and still developing. Formal policy can sit apart from classroom practice, a pattern institutional theory explains through the decoupling of formal structures from everyday organizational activity (Meyer & Rowan, 1977). Higher education is especially susceptible to such gaps when goals are ambiguous, expertise is distributed, environments are changing, and responsibility is contested (Kezar & Eckel, 2002). Recent research suggests that AI governance is developing under many of these same conditions. One study of professional practice in a heavily regulated setting finds that professionals respond to AI disruption through forms of interactive governance shaped simultaneously by institutional, organizational, and social influences (Chen et al., 2025), while broad expectations for AI adoption without precise prescriptions produce uneven faculty responses (Chan & Colloton, 2024). Together, this work helps explain why institutional direction may coexist with substantial variation in what faculty ultimately do in their classrooms.

Three interpretations of institutional inaction are possible. Silence may represent a deliberate effort to preserve flexibility or postpone conflict, consistent with the idea that non-decision-making can itself function as a form of decision-making (Bachrach & Baratz, 1962; Matland, 1995). It may also reflect traditions of academic freedom that leave substantial pedagogical authority with faculty (Tierney, 2004). A third possibility is that policy simply lags behind technological change, particularly when AI is treated primarily as a technology, risk, or compliance issue while instructional questions remain unresolved. This dissertation does not presume which explanation is correct. Instead, it examines the consequence of uncertain policy for faculty who must make course-level AI decisions without fully specified guidance.

Why might faculty in different fields interpret AI differently? Becher and Trowler (2001), building on Biglan (1973), classify academic fields along two dimensions: whether knowledge is hard or soft and whether the field is pure or applied. Disciplines function as epistemic cultures, carrying shared assumptions about evidence, rigor, authorship, method, and the legitimacy of particular tools. These norms can shape how AI is evaluated. Hard pure fields may place greater emphasis on verifiable method and reproducible reasoning, while soft applied fields may give more weight to professional judgment, ethics, and situated practice. Research predating genAI shows that disciplinary context and institutional commitment can shape faculty technology use (Waggoner, 1994; Shelton, 2014), while recent studies report disciplinary differences in AI orientation, with STEM and applied fields often showing greater openness and non-STEM fields more frequently emphasizing ethical or cultural concerns (Yang et al., 2026). Becher and Trowler therefore provide a structure for examining patterned disciplinary differences without assuming that individual faculty within a field will respond uniformly.

Weick’s sensemaking theory explains how faculty interpret unsettled institutional conditions and determine how to act. Sensemaking involves constructing plausible understandings that allow action to proceed despite ambiguity (Louis, 1980; Weick, 1995; Weick, Sutcliffe, & Obstfeld, 2005). In the context of AI, faculty confront contested benefits, discipline-specific risks, and still-developing standards for appropriate use. Their judgments are shaped by prior experience, disciplinary identity, peer interaction, and available institutional cues. Sensemaking is therefore social and situated rather than purely individual, which helps explain why the same institutional guidance can lead to different classroom decisions even within a single department. It also does not imply movement toward adoption, since integration, restriction, prohibition, and continuity with existing practice may each represent a coherent response to ambiguity.

The focus on interpretation under ambiguity also connects to the subjective tradition in educational management associated with Greenfield (1973) and later Bush (2003), which emphasizes that organizational reality is constructed through the meanings members attach to their circumstances. Its limitation for this study is that it offers less leverage for explaining why interpretations may vary systematically across disciplinary settings or how those interpretations become consequential classroom rules. The framework developed here extends that tradition by locating faculty within disciplinary cultures through Becher and Trowler, explaining interpretation through Weick, and tracing discretionary implementation through Lipsky. Loose coupling in educational organizations further creates the local space in which such interpretation and discretion can operate (Weick, 1976).

Governance conditions, disciplinary norms, and individual interpretations become consequential for students when faculty decide what their courses require. Lipsky’s (2010) street-level bureaucracy framework explains this point of implementation: front-line professionals exercise judgment under conditions of ambiguity, limited resources, and incomplete rules, and those judgments help determine how policy is experienced in practice. For students, AI governance is encountered through syllabi, assignment requirements, grading practices, and instructor explanations, and the expectations recorded there vary with the discretion their instructors exercise.

Faculty fit this framework with some modification. Academic freedom, disciplinary expertise, and shared governance make professional discretion an expected feature of academic work (Lovell, 2023; Khelifi, 2025). Grading offers a useful comparison as institutions require grades and establish academic standards, yet they do not prescribe what an A must mean in every course. Faculty exercise professional judgment within those broader boundaries and AI decisions operate through a similar discretionary channel: bounded by institutional expectations, interpreted through professional expertise, and consequential for students.

The three frameworks do different work at cultural, interpretive, and implementation levels. Becher and Trowler locate faculty within disciplinary cultures. Weick frames sensemaking as the interpretation of ambiguity into a workable understanding. Lipsky frames discretion as the organizational condition under which faculty classroom decisions carry consequence for students. Together, the three identify relationships among disciplinary context, reported interpretation, and discretionary practice, and they do not trace a sequence from policy to classroom.

The study design follows from this framework. Disciplinary placement is established through the survey distribution structure and organized within the four Becher and Trowler quadrants. Reported orientations toward ambiguity, peer interpretation, disciplinary norms, retrospection, and the plausibility of competing readings provide the evidence of sensemaking, while case-by-case judgment, variation across courses, adjustment within a term, adaptation of guidance, and gap-filling provide the evidence of discretionary practice. Public course documents add a second perspective by showing whether AI is prohibited, permitted, conditionally allowed, subject to disclosure, left unaddressed, or governed through template language. Comparing those materials with faculty reports shows where the two sources describe similar or different conditions, and where they describe the same condition from different vantage points. The aim is not to judge compliance or to reconstruct deliberation. It is to examine how disciplinary context and reported interpretation relate to the classroom conditions students meet.

The study is a two-phase quantitative design. Phase 1 is a completed documentary analysis of student-facing generative AI governance in public course documents, and Phase 2 is a faculty survey examining the correlates of reported discretionary AI practice. The two phases are sequential and address different units of analysis: Phase 1 records what governance is visible across institutions and disciplines, and Phase 2 examines faculty-level relationships within one of those institutions. The phases are not parallel measurements of the same construct and are not used to validate one another (Creswell & Creswell, 2023).

The documentary analysis draws on publicly available HB 2504 syllabi and is substantially complete, while the faculty survey will proceed following proposal approval and IRB clearance. Phase 1 establishes the empirical pattern that motivates the faculty-level question. Phase 2 carries the study's theoretical weight and supplies its central quantitative evidence: the closed items provide the analytic measures for RQ2 and RQ3. The instrument collects no qualitative responses, and no part of the present study is analyzed qualitatively.

The documentary analysis examines 300 upper-division undergraduate syllabi from five Texas public R1 flagship institutions across twelve disciplines, three in each Becher and Trowler quadrant, with five syllabi per institution-by-discipline cell. Eligibility required a complete public syllabus, upper-division status under each institution’s numbering convention, membership in the discipline-prefix crosswalk, and a Fall 2025 or Spring 2026 term. Selection was outcome-blind and deterministic. Within each cell, eligible records were ordered by prefix, catalog number, term, section, and source URL before AI governance outcomes were examined, and the first five eligible records were retained. Each retained row preserves its source URL, original document, extracted text, and document and text hashes, and an independent verification script rechecks retained hashes, matrix balance, counts, and exact tests.

Coding first distinguished explicit AI language from an operative student-facing AI-use rule. A syllabus could mention AI as course content without stating what students may or may not do. Visible rules were classified as Full Prohibition, Partial Restriction, Disclosure Required, Full Permission, or Scaffolded Integration, while AI-Silent indicated that no operative rule was visible. Two independent coders applied the same codebook to a blind 60-syllabus subset, one from each institution-by-discipline cell. Agreement was 59 of 60 for AI-language presence (κ = .963), 50 of 60 for governance position (κ = .773), and 54 of 60 for course-specific rationale (κ = .833). Among the 38 blind-subset rows classified as governance-visible, rationale agreement was 32 of 38 (κ = .582), a weaker and less precise result that is reported conservatively and interpreted cautiously.

The completed analysis found visible student-facing rules in 163 syllabi (54.3%) and no visible rule in 137. Among governance-visible syllabi, 49 used Full Prohibition, 84 Partial Restriction, 20 Disclosure Required, 3 Full Permission, and 7 Scaffolded Integration. The clearest disciplinary difference was between hard and soft fields: visible governance appeared in 56 of 150 hard-field syllabi (37.3%) and 107 of 150 soft-field syllabi (71.3%), with an exact discipline-profile test of p = .006 and a Bonferroni-adjusted p = .013. Pure and applied fields showed little difference (52.7% versus 56.0%, p = .851). The four-quadrant comparison was weaker and is treated as exploratory (p = .045).

The evidentiary boundary remains constant. The documentary record supports claims about visible public course governance, not instructor motive, enforcement, or actual student use. Template and source-system cautions also apply. Repeated syllabus language may reflect institutional templates rather than independent faculty reasoning, and a missing statement may reflect a practice considered settled enough to go unwritten.

The researcher holds an Assistant Provost role at the study institution. That position provides institutional knowledge relevant to the research problem while also creating the possibility that prior experience and organizational familiarity could shape interpretation. Recruitment and data collection will be separated from the researcher's administrative responsibilities, and the survey will not collect unnecessary identifying information. Analytic decisions and assumptions that could materially affect interpretation will be documented as needed, and the Phase 1 documentary findings and Phase 2 survey findings will be reported separately and compared where the two phases describe the same condition.

The site is a large public research university whose scale, disciplinary range, and decentralized governance make faculty-level interpretation visible. All four disciplinary quadrants are represented within a common institutional governance environment, allowing disciplinary variation to be examined while holding the institutional setting constant. The population includes faculty with instructional or course-design responsibility, including tenured, tenure-track, non-tenure-track, and adjunct instructors. Research-only faculty, postdoctoral fellows, and visiting faculty without instructional responsibility are excluded.

The Qualtrics instrument contains twenty-one respondent-facing survey items: five background, classification, and control items and sixteen analytic items.

Fourteen analytic items contribute to three planned composite measures: institutional guidance conditions with four items, sensemaking orientation with five items, and discretionary classroom practice with five items. Two additional analytic items are reported descriptively and do not enter a composite: one records the clarity of expectations at the department or program level, and the other records whether faculty see the concerns raised by AI as different in their field than in other fields. Closed indicator items use a five-point agreement scale from strongly disagree to strongly agree, allowing a neutral response where faculty have not formed a stable position.

Composite use is contingent on pre-specified reliability and item-total diagnostics rather than assumed in advance. If the items do not support defensible aggregation, narrower item-level analyses will be reported as exploratory and will not be treated as equivalent tests of the construct-level RQ3. Because the sensemaking and discretionary-practice items represent theoretically related but distinct facets of their respective constructs, dimensionality will be examined before aggregation, and a weak internal-consistency estimate will not on its own be read as evidence that a construct failed.

The instrument does not treat sensemaking as a validated psychometric scale. Instead, it examines reported orientations and classroom responses through sensemaking theory, with connections to Weick's properties examined rather than assumed. The instrument collects closed-item responses only and includes no open-ended prompts. The decision to collect no qualitative data in this study reflects the two-phase quantitative design, in which the process question is reserved for a subsequent study.

At the end of the survey, respondents may choose to open a separate, unlinked Qualtrics form to provide an email address for a later follow-up study. No survey response ID or other identifier will be passed to that form, and the contact information will not be used in any analysis of the present study. Qualtrics may record limited administrative information depending on distribution settings, but any such field will be documented, and no undeclared field will enter an analysis.

Data collection will proceed through instrument finalization, IRB review, and target mid-semester deployment. Exempt review is the anticipated route because the analytic survey is anonymous, adult, minimal-risk, and designed without direct identifiers. The administrative role will be disclosed in the IRB application, and college administration will not serve as recruitment intermediaries unless an approved neutral distribution process permits it. The eligible invitation pool will be established from faculty with current instructional or course-design responsibility. Because the study compares disciplinary quadrants, the strength of subgroup claims will depend on achieved cell sizes rather than the aggregate response count.

Phase 1 analysis is complete and is reported as such. The documentary data are analyzed by discipline and institution, with exact tests for the discipline profile and quadrant contrasts and Bonferroni adjustment for the multiple comparisons involved. Phase 2 analysis will be descriptive and comparative for RQ2, while RQ3 will provide the study's primary multivariable analysis. RQ2 will summarize item-level and composite distributions across disciplinary contexts, with quadrant comparisons and effect sizes reported where achieved cell sizes support them.

RQ3 will model discretionary classroom practice as the outcome, with sensemaking orientation as the focal predictor and perceived guidance conditions, disciplinary context, teaching experience, AI familiarity, and course load included as covariates. Ordinary least squares regression with robust standard errors will be used if the outcome supports defensible continuous treatment; ordinal logistic regression will be used if it is retained as an ordered outcome.

Composite scores will be used only if the pre-specified reliability diagnostics support aggregation. For discretionary practice, respondents teaching one course may select a not-applicable response on the across-course variation item; that response will be treated as missing, and the composite will be calculated only when at least four of the five discretion items are available.

The disciplinary moderation analysis is pre-specified as exploratory rather than as a research question. Interaction terms will be added to the RQ3 model only as an exploratory check, and the minimum detectable interaction will be reported so that a null result is not interpreted as evidence against moderation.

Respondents who cannot be classified confidently into one of the four Becher and Trowler quadrants will be retained as interdisciplinary or uncertain for descriptive purposes rather than treated as a fifth theoretical quadrant; sensitivity analyses will document how their treatment affects the estimates. Analyses will use available responses, with the extent of missing data reported, and the assumptions of the fitted models will be examined and reported.

The phases are reported in sequence. Phase 1 reports the visible pattern of student-facing AI governance across institutions and disciplines, and Phase 2 reports faculty-level variation and the association examined in RQ3. The two phases address different units of analysis and different samples, so their results are presented side by side and are not combined into a single estimate. Phase 1 identifies the disciplinary pattern that motivates the faculty-level question, and Phase 2 tests whether reported orientations are associated with reported discretionary practice within the institutional setting Phase 1 described. The limitations of each phase are stated separately and are not treated as equivalent.

The study has several limitations. The faculty survey will be conducted at one institution, so its findings are bounded to that governance context. The documentary analysis is limited to public course materials and cannot establish instructor motive, enforcement, or actual student behavior.

Survey responses reflect retrospective accounts of practice, and the closed items capture reported orientations and practices with a limited number of indicators per construct. For that reason, sensemaking is treated as an interpretive lens rather than a directly measured psychological construct. The study therefore describes what faculty report about interpreting and implementing AI guidance. It does not observe the process by which guidance reaches the classroom, and its findings cannot support claims about how that process unfolds. Because the instrument collects no qualitative data, the study also cannot explain why a reported association holds or fails to hold; that question is reserved for the follow-up study.

Finally, the study captures one period of rapid AI policy development. Nonresponse, uneven participation across disciplinary quadrants, and changes in institutional guidance may affect the patterns observed. Findings will therefore be interpreted as bounded evidence of faculty-reported sensemaking and discretionary practice rather than as causal or broadly generalizable conclusions.

This candidacy proposal establishes the need, framework, and design for examining the variation in classroom AI governance that students encounter and the faculty-reported conditions associated with it. Institutions, systems, teaching centers, and instructors are already producing AI guidance, and students meet that guidance unevenly across courses. Together, Becher and Trowler, Weick, and Lipsky frame disciplinary context, reported sensemaking, and discretion as related dimensions of that governance. The study proceeds in two quantitative phases. Phase 1 reports the student-facing governance visible in public course documents, establishing the disciplinary pattern that motivates the faculty-level question. Phase 2 examines what faculty at the study institution report about their orientations, the guidance they perceive, and their classroom decisions, and tests whether reported orientations are associated with reported discretionary practice after accounting for disciplinary context and professional characteristics.

Together, the phases can describe the range of reported faculty practice and indicate where institutions might improve transparency and support without flattening disciplinary differences or treating faculty discretion as a problem to eliminate. The process by which faculty arrive at a specific classroom decision is not observed in this design. That question is well suited to a subsequent qualitative study, which the separate optional recontact form is designed to make possible.