The departmental adoption kit

For chairs, program directors, and anyone standing up a department’s AI practice.

Most universities now leave instructional AI decisions to the individual course. That protects academic freedom, and it leaves students to reconstruct the rules course by course, five syllabi with four different policies in one term. The department is the layer that can fix this without touching anyone’s freedom. A department that adopts one vocabulary, asks every course to state its policy, and reviews those policies for clarity gives its students coherence at almost no cost to its faculty.

This kit is the day-one toolset for that work: a model departmental policy statement, the shared vocabulary, the four assignment categories, the five questions every syllabus should answer, and a one-semester rollout sequence. Everything here is meant to be copied, cut, and adapted. The reasoning behind each piece is in Part III of Learning with AI; this kit is the part you paste.

Two companions sit beside it in this folder. The policy self-audit maps what your institution already covers, what your department covers, and what falls to nobody yet; run it first if you want the gap list before you build. The regulatory landscape note tracks where AI use crosses from policy into law.

The rule that holds it together

Require every course to make its AI policy clear. Do not require every course to have the same policy.

A calculus course, a proof course, and a modeling course will draw the lines in different places, and they should. What students need is not sameness but legibility: the same words meaning the same things, stated where students can find them. Clarity is the department’s standard. The lines themselves belong to the instructor.

A model departmental policy statement

Paste this into the department handbook or website and adapt the bracketed parts.

The department recognizes that AI tools are now part of professional and scholarly work in [discipline]. Our courses teach students to use these tools well while building the independent knowledge, skill, and judgment the degree certifies.

Every course in the department states in its syllabus how AI may and may not be used. Assignments are labeled by the role AI may play: no AI, AI as tutor, AI as collaborator, or AI as object of study. Students follow the rules of each course, verify any AI-assisted work they submit, protect private and restricted information, and disclose AI use where the course requires it.

Instructors own grading, feedback, and academic-integrity judgments. Concerns about misuse are handled through process evidence, conversation with the student, and [the university’s academic-integrity procedure]. Detector scores are not treated as the sole or primary evidence that misconduct occurred.

Three notes on adapting it. Name the discipline, because a statement about “AI in statistics” lands harder with statistics faculty than a generic one. Align the last paragraph with your institution’s actual integrity procedure. And date the statement, with the next review on the calendar; an undated policy cannot be maintained, and in a field moving this fast an undated policy is an expired one.

The shared vocabulary

Words fail across courses before policies do. These terms, used consistently in syllabi, assignment sheets, and department documents, are most of the coherence students will ever notice.

Three things a course certifies. Core competence is what a student can do with no AI at all, or with only narrow, specified help. AI-assisted practice is what they can do with AI as a permitted tool while keeping control of the work. Trustworthiness is how well they verify, disclose, and take ownership of AI-assisted work. Courses should say which assignments measure which, because a student can be strong in one and weak in another, and a single grade hides that.

Kinds of work. No-AI work is done without AI help, apart from accommodations or tools the instructor has approved. Bounded-AI work allows limited, specified help: one approved tool, a restricted kind of prompt, a fixed window. Open-AI work allows broad use, subject to disclosure and verification. AI-assisted work is improved with AI but written, checked, and owned by the student. AI-generated output is whatever the system produced: text, code, solutions, summaries, analyses.

Practices. Disclosure is a brief factual statement of what tools were used, what they contributed, and how the student checked the result. Verification is that checking: against course materials, sources, data, tests, or the student’s own reasoning. Process evidence is the trail that shows how work was made: plans, drafts, prompt logs, version histories, verification notes. An oral defense is a live explanation of the work. A random audit is a routine spot check in which a student walks through submitted work; it is a normal part of assessment, not an accusation.

The four assignment categories

Every assignment names the role AI may play. Four categories are enough, and they are simple enough for a syllabus and flexible enough for any discipline.

NAI: No AI. AI use is not allowed. These assignments measure what a student can do alone: closed-book exams, in-class writing, oral explanations, live coding checks, foundational calculations. They serve a specific purpose and should not dominate a course.

AIT: AI as Tutor. AI may explain, quiz, and coach, but the submitted work is the student’s own. A student might ask for practice problems or a second explanation of a concept before an exam, then answer unaided.

AIC: AI as Collaborator. AI may help brainstorm, draft, revise, debug, critique, or test. The student discloses the use and verifies what matters. This is the category for work that mirrors professional practice: coding, modeling, research, design, revision.

AIS: AI as Object of Study. Students analyze the AI itself: find its errors, critique its bias, compare tools, repair its reasoning, evaluate its code. These assignments make the AI the thing under examination instead of the thing doing the work.

The boundaries are not always sharp. A student who asks for an explanation and then drafts from it has moved from tutor to collaborator in one sitting. The label names the dominant role in producing the submitted work; where a task genuinely mixes roles, the assignment should say so.

Five questions every syllabus answers

Ask each course to answer these, in writing, where students will see them.

  1. What AI use is allowed in this course, and what is off limits?
  2. Which assignments are NAI, AIT, AIC, and AIS?
  3. What should a student be able to do with no AI help at all?
  4. How should students disclose AI use?
  5. What materials may never be uploaded into an AI system?

Large and multi-section courses should answer a sixth: how the course checks understanding at scale. Short no-AI quizzes, process logs, random oral audits, in-class writing, and draft histories all work, and naming the method in the syllabus makes the spot check routine instead of adversarial.

For question 4, point students at the disclosure templates in this repository rather than inventing a format per course; a disclosure should read like a note on method, not a confession. For question 5, the department should echo the institution’s floor: no private information about people, no unpublished research data without permission, no secure exam materials, no confidential documents.

Standing it up in one semester

  1. One meeting to adopt the vocabulary and the four categories. This is the only department-wide decision the kit requires. Skeptics do not need to change how they teach; they need only label what they already do. In the same meeting, name the courses that treat AI literacy as a learning outcome in its own right; those are the natural homes for AIS work.
  2. Every course answers the five questions in its syllabus and labels its assignments.
  3. Collect the statements and review them for clarity, not sameness. The review question is whether a student could follow the rules as written, and whether the rules are fair.
  4. Hold one calibration conversation. What does verification mean in this discipline? Where process evidence is graded, calibrate the TAs who grade it.
  5. Support the largest courses first. They carry most of the students and most of the risk, and they are where clarity pays off soonest. Alongside them, adapt two or three of the repository’s assignment templates into discipline-specific samples the whole department can borrow from.
  6. At term’s end, review and date. Collect what confused students, update the statement and the samples, stamp the revision date, and put the next review on the calendar.

A department that runs this sequence once has something most institutions do not: a documented, tested, coherent practice. That is worth more than it looks. When the faculty senate or the provost’s office finally takes up AI policy, it will look for working examples, and departments that have one shape the outcome. Departments that do not become subject to whatever is decided without them. When that conversation arrives, the book’s model institutional policy statement in Part III is the ready starting text for the university-level version of what your department already practices.

What to hand your faculty

The instructor materials in this repository do the course-level work so the department does not have to write it.

For students, the parallel set: start here, the disclosure templates, and the student compact, which some departments adopt as a shared signing document.

Where this comes from

This kit adapts Part III of Learning with AI: A Framework for Students, Instructors, and Universities (James M. Hyman, SIAM Books) for standalone departmental use. The book carries the reasoning, the evidence, and the institutional layers above the department: governance, training plans, implementation, evaluation metrics, and the risk register.

Part of the companion repository for Learning with AI. These materials are free to use and adapt with attribution.