AI Policy Lab: Facilitator Workbook
For the person running the Lab. Participants use the workbook. Fill in the Prepare tab once for each run, and the Run and Reference tabs, the printed run sheet, and the participant handout carry those choices.
No run set up yet. Open Prepare to begin.
Prepare
The choices below set up a run. This browser saves them as you type. Change the format and the Run tab re-times every block.
Before the first session
Email to participants
Filled from the fields above. Copy it into your mail client; Garamond is applied there.
Run
One block per module or activity, in order. Each block opens to the same six lines: say, do, watch for, alternatives, short on time, slides. Start a block's clock when you begin it.
Reference
What belongs to no single block: the procedure and the three questions, the five principles, the cases, the questions participants ask, and the sources.
The procedure, with the question added in 2026
- Identify the genre. What kind of document is this, issued by whom, to whom, under what authority? Genre sets what the document is able to say.
- Locate the glitch. Where does the document strain against itself: a contradiction, a hedge, a term invoked and never defined, course-level language in a university document, a trace of revision? Where two documents conflict, decide whether both are current (a contradiction) or one has replaced the other (a supersession). A replaced document is a version note.
- Read the reveal. What does the glitch make legible about the institution's commitments, assumptions, or fears? This is a reading, and no verdict on the institution.
- Name the abstraction. Which general claim holds the document together, and where does it fail? "Faculty discretion" moves risk without governance. "Responsible use" names individual conduct while structural conditions go unaddressed. "Integrity" frames AI as a cheating problem.
The three questions to ask of any policy
- Conceptual: what does this policy name, and is the thing it names defined?
- Structural: whose problem does this policy make AI, and which conditions (procurement, data, infrastructure) stay out of view?
- Operative: what judgment does the policy route, to whom, and with what support?
The five principles
- Scope terms by function. "Tools that generate text, code, images, or data from prompts," since a product name goes out of date with the product.
- Date the policy. "This policy applies to Spring 2027 and will be reviewed before the following semester."
- Name the revision trigger. "If a materially different AI capability becomes available, or the university publishes a policy, this policy will be updated."
- Separate durable principles from contingent prohibitions. The commitment (what the course develops and assesses) stands apart from the rule (which functions are barred this term), because different bodies write each and the first outlasts the second.
- Explain the reasoning. Students comply more with policies they understand, and a reviewer knows what to keep when the tools change.
Cases on record
NC State, the record in the censusEleven documents, seven verified, three announced with no text, one promise since removed from its page. The OIT approved list clears Gemini, Copilot, and ChatGPT for Red data while a November 2024 OIT notice limited them to green and yellow. The AI Hub bars university data from tools OIT has not vetted; its advisory group has posted a charge and 2026 minutes and no report. The DELTA course AI guidelines page states that as of May 2026 there is no official university policy regarding AI use, beside three sample syllabus statements. The Code of Student Conduct (REG 11.35.05, revised July 2025) contains no AI language; its integrity provisions rest on the unauthorized-assistance clause (8.1.b) and the Pack Pledge. The Graduate Handbook (3.6) requires theses and dissertations to document generative AI use, the one formal instrument naming AI.
NC State against twelve UNC System peersWhere the highest-order document sits: four peers govern AI through a provost policy or guideline (Georgia, Maryland, Michigan State, Minnesota), two place the rule in the system office (Illinois, Texas A&M), UC Davis divides it between campus IT and the UC Office of the President, Arizona built an Office of Responsible AI under a Chief AI Officer, Virginia Tech and Purdue anchor in an IT or working-group framework, and Penn State sits with a vice provost. NC State is the one institution in the set whose highest-order document is committee guidance, with no provost, chancellor, or system instrument above it. The conduct code: Ohio State and Texas A&M wrote AI into the integrity code, nine of the thirteen (NC State among them) have a code with no AI language, and two (UC Davis, Maryland) have codes the census was unable to read. Minnesota closes the gap by interpretation, a provost memo applying the existing clause on unauthorized use of online learning support and testing platforms to AI tools. Contradictions: nine of thirteen carry at least one, and the recurring type is NC State's own, an IT or licensing page and a teaching or college page disagreeing on what a licensed tool permits (Illinois, Ohio State, Arizona, Michigan State). Announced bodies without reports: UC Davis four, NC State three, including the Spring 2026 Update and Pilot Report the March 27 minutes cite. Vendors: every peer names at least two licensed products, and three run an institution-branded wrapper (TerpAI, Texas A&M AI Chat, HokieAI). The land-grant peers with a state system above them inherit a system rule they then adapt; the UNC System has issued nothing comparable in the census so far, which leaves NC State's governing layer thinner than the land-grant norm by one tier.
Questions participants ask
- What should the policy say?
- The what follows from the why. The workbook's Part IV holds template language at three levels, and template language without a stated rationale is what the procedure finds glitches in. Work through the three questions first, then the templates.
- Is AI use in my class cheating?
- At NC State, the Code defines cheating as using assistance the instructor has not authorized (8.1.b), so the answer sits in the syllabus. Conceptually: which function? Structurally: what the assignment is doing and what AI does to it. Operatively: who decides, on what grounds, and what recourse a student has.
- What does NC State require of us?
- No university-wide AI policy has been published. The AI Hub bars university data from unvetted tools, OIT clears tools by data classification, the Graduate School requires thesis disclosure, and the Code leaves authorization to the instructor. The AI Advisory Group has posted a charge and minutes; the AI Hub Learning, Curriculum, and AI Literacy working group is the venue to watch for curriculum.
- Will we be required to put a statement in every syllabus?
- Nothing on record at NC State requires it. Among the twelve peers, Michigan State requires one by provost guideline and Minnesota applies an existing code clause to AI by provost memo; the others govern through system offices, IT frameworks, or a dedicated office. The question for the unit is which instrument, if any, it wants, and at which level.
- Should we use detection software?
- Run the three questions. Conceptually, a detector reports a probability score, which establishes nothing about what a student did. Structurally, Liang and colleagues (2023) found a 61.3 percent average false-positive rate for non-native English writers across seven detectors, and Weber-Wulff and colleagues (2023) found none of fourteen tools reliable. Operatively, Penn State's integrity office states the position in one sentence: detector results are not reliable enough to use as the basis for an academic decision. CWSP does not use them, and the reasons are on the framework page.
- Why "glitch"?
- The term names a productive failure, a place where language reveals more than it intends, and it comes from the theoretical frame behind the Lab. "Tension" or "gap" substitutes without losing the analysis if the room finds it too playful.
Sources
- Cole, K. K. (2026). AI Governance Census of US Colleges and Universities, version 0.5. https://doi.org/10.5281/zenodo.23887259 (opens in a new window) (concept DOI, resolves to the latest version).
- Cole, K. K. (2026). AI literacy is not one thing: Conceptual, structural, and operative levels for higher education governance. Pedagogies: An International Journal.
- Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., and Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4, 100779. https://doi.org/10.1016/j.patter.2023.100779 (opens in a new window)
- Weber-Wulff, D., Anohina-Naumeca, A., Bjelobaba, S., Foltýnek, T., Guerrero-Dib, J., Popoola, O., Šigut, P., and Waddington, L. (2023). Testing of detection tools for AI-generated text. International Journal for Educational Integrity, 19(1), 26. https://doi.org/10.1007/s40979-023-00146-z (opens in a new window)
- North Carolina State University. (2025). REG 11.35.05, Code of Student Conduct. Last revised July 22, 2025.