AI Policy Just Moved Down to Your Department. The Guideline Names Three Kinds of AI.

Mississippi asked every department for an AI policy, and the guideline it handed them names three kinds of AI. Most syllabus policies name one.

For three years the AI policy question in higher education has been an institutional question. A provost’s office writes a statement. A committee reviews it. It lands on a webpage under a heading like Responsible Use, and it says something careful about academic integrity and generative tools. Then it sits there, because a sentence written at the altitude of a whole university cannot tell a studio art instructor and a nursing clinical coordinator the same useful thing on the same day.

The University of Mississippi has now done the obvious next thing. It pushed the writing down to the departments. And the obvious next thing turns out to be much larger than it sounds.

What actually happened, and when it lands

According to reporting in The Daily Mississippian on August 23, every academic department at the university will introduce its own artificial intelligence policy this semester, and those policies are expected to appear in every course syllabus. Departments were told to write them last spring. The university’s AI Task Force supplied the guidance about what each policy has to cover.

That guidance is the part worth reading twice. The Daily Mississippian quotes the AI Guidelines for Teaching and Learning directly: each academic department or program is responsible for developing artificial intelligence guidelines that address student use of generative AI in course components, address student use of agentic AI in course components, address student use of assistive AI in course components, and hold students accountable for infractions.

Read that list slowly. It is not one rule about AI. It is three separate rules about three separate categories of tool, applied not to the course in general but to course components. Assignments. Discussions. Exams. Labs. Whatever the parts of your course happen to be.

Robert Cummings, who chairs the university’s AI Task Force, spent the summer meeting with administrators, deans and chairs so that departments understood the principles underneath the requirement. He described the situation without pretending it is settled. “There is a natural tension between groups of faculty who are resisting AI and groups of faculty who are embracing AI,” he told the paper. “Both of those perspectives, in my opinion, are correct, and that’s what makes this so challenging and so interesting.”

Three categories, and only one of them is in your current policy

Almost every syllabus AI policy written since 2023 is a policy about generative AI. It talks about text a student did not write and images a student did not make. That is one column of three.

Generative is the familiar one. A student asks for a draft and receives a draft. Your existing sentence probably covers it, whether the sentence permits, restricts or forbids.

Assistive is the one already running in your students’ toolbars whether anyone wrote a rule or not. Spelling and grammar correction. Read aloud. Live captioning. Translation. Predictive text. Reading level adjustment. This is the column where a blanket prohibition quietly collides with accessibility, because for some students in your room these are not conveniences, they are the accommodation. A policy that bans AI without naming this column has just written something it cannot enforce and would not want to.

Agentic is the one with no precedent, and it is the reason this guideline matters beyond one campus. An agent does not hand a student a paragraph. It takes actions. It opens the sources, runs the code, fills the spreadsheet, submits the form, and reports back on what it did. The student may have set the goal and never touched a step. Nothing in a policy about text a student did not write reaches that, because the issue is not authorship of a product. It is authorship of a process.

Which means the hardest sentence in your new department policy is about a category of tool most syllabi have never had to name, and there is no template for it in the folder where you keep last year’s syllabus.

The clause that sets the real workload

The guidance also asks for annual renewal, so that policies keep pace with the tools. That single word, annual, is what converts this from a task into a standing obligation.

Run the arithmetic on your own load. Three categories. Times the number of graded components in one course, which for most of us is somewhere between four and a dozen. Times the number of distinct preparations you teach. Times every term, forever, because the category list itself will move. Agentic was not a word in a syllabus two years ago. There will be a fourth column before there is a fifth syllabus.

This is why the department level is the right altitude and still not a small ask. Rock, quoted for the School of Education in the same report, put the tension plainly. Departments inside one school can share a purpose and still need different rules. “We have a Department of Higher Education preparing professionals to work in higher education, while other departments focus on areas such as teacher education and early childhood education,” she said. “The goal of teaching and learning is the same across those areas, but how AI is used and what guidelines or requirements are needed can be very different.”

Cummings reached for a better image than any policy memo has managed. “A university is like a city, and in a city, you have different neighborhoods, and in different neighborhoods, you’ve got different approaches to solving problems and moving forward,” he said. Students “are going to be in different departments, and they’ll have different approaches, but wherever they are, they’re owed clear guidelines in the classroom.”

Owed is the right verb. A student who reads four syllabi and finds four silences has not been given a policy. They have been given a guess, and they will guess in the direction of whatever their roommate got away with.

What the faculty skill in the third column actually is

There is a temptation to treat the agentic column as an enforcement problem and write it as a prohibition. That is the wrong instinct, and not for a soft reason.

The emerging frameworks for AI fluency in education do not describe the relevant competence as detection. They describe it as managing the actions of a system: delegating a task responsibly, setting the parameters before it runs, and reviewing what came back. That is a teachable skill with observable evidence, and it is one you already assess in other forms. You do it every time you evaluate whether a student chose the right method before you evaluate whether they got the right answer.

So the agentic sentence in a good policy probably does not read as a ban. It reads as a disclosure requirement plus a review requirement. What did you delegate. What did you check. Show me the step you did not automate.

The practical objection to all of this is time, and it is a fair objection. In a Gamma session for higher education, an instructor who serves on his own university’s AI task force said the thing that actually moves faculty is a low bar for entry and a familiar format, not a new system to learn. He is right, and it applies here. The department that ships a usable policy this semester will be the one that started from a document faculty already own rather than from a blank page.

Four moves, this week

You do not need your department’s final language to start, and waiting for it is how the deadline arrives with nothing written.

One. Take a single assignment, not the whole course. Write three sentences for it, one per category. Notice which sentence took the longest. It will be the third one, and that tells you where the department conversation is going to stall.

Two. Read your existing syllabus policy and mark which of the three columns it actually addresses. Most instructors find one and a half.

Three. Check the assistive column against your accessibility statement, on the same page, side by side. If they contradict each other, the accessibility statement wins, and your AI paragraph has to be rewritten to say so out loud.

Four. Bring the one worked assignment to the department meeting instead of an opinion. A colleague with a finished example changes a meeting. A colleague with a position extends one.

The part nobody has written yet

Every policy on this list governs what a student may do. Not one of the three columns says a word about what the instructor may delegate to the same three kinds of tool, in the same course, on the same work.

You will be asked that question this year, probably by a student, probably in week four, probably about grading. The department that has already written its own answer will sound like it meant it.

When you are ready to move from one assignment to the whole course, the Course Refresh Assistant takes an old syllabus and gives you back a live Course Record in minutes, so the three columns land in every component instead of only the one you had time for. Do the first one by hand anyway. That is how you will know whether the rest are right.