The most important generative-AI policy at a university may not be the one published on its website. It may be the sentence written at the top of an assignment. Students now move between courses in which AI can mean completely different things. One instructor may encourage it for brainstorming. Another may permit grammar assistance but not generated prose. A third may prohibit it entirely.
Those differences can be educationally justified. The problem begins when the differences are invisible. Universities across South Asia should require every substantial assignment to carry a short AI boundary. The boundary could use simple categories: no generative AI; brainstorming only; revision only; or permitted with disclosure and verification. The wording can vary by discipline. The principle should not. Students should know the rule before they begin.
This would improve more than compliance. It would force instructors to connect AI rules to learning goals. If an essay is designed to assess a student’s ability to construct an argument independently, the instructor can explain why the first draft must be unaided. If an assignment is designed to teach critical AI use, the instructor can permit the tool while requiring students to verify consequential factual claims and explain important decisions. That is more useful than a blanket statement telling students to “use AI responsibly.”
Responsible for what? A student needs to know which intellectual work cannot be delegated. Assignment-level boundaries also reduce the temptation to treat AI policy as detective work after submission. Trying to reconstruct a student’s entire composing process from polished final prose is difficult and can create an adversarial relationship between teacher and student. A visible rule changes the conversation. The teacher defines expectations first. The student can then document the process that matters.
For some assignments, a short human first attempt could be part of that process: an outline, a paragraph, a list of questions, or a rough explanation of the problem before AI assistance begins. The goal is not surveillance. It is to preserve evidence of learning. South Asian institutions should also think about access.
Students do not all use the same devices, internet connections, or AI subscriptions. If an instructor requires AI use, the university should make sure students can access the necessary tool. If AI is optional, students should not be disadvantaged because somebody else can afford a more capable paid system. The assignment should assess the intended skill, not the purchasing power behind the software.
There is no need for universities to agree on one universal rule. A literature seminar, programming course, and research-methods class may reasonably need different boundaries. But ambiguity should not be mistaken for flexibility. The tools will continue changing. Product names and features will come and go.
Four questions will remain useful: What ability is this assignment trying to develop? What work must the student perform? What assistance is permitted? Who is responsible for the final result?
Put the answers where students actually need them: on the assignment itself. The future of academic integrity will not be secured by making AI policy longer. It will be secured by making expectations clearer.
Swaansh Mahat
Swaansh Mahat is a fourth-year BA student majoring in English at Tribhuvan University and an independent writer based in Kathmandu, Nepal. He writes on youth, education, technology and authorship in the age of generative AI, with opinion pieces published in national and international publications.
