The integrity question, reframed
When generative AI became widely available, the first reaction in many universities was defensive: ban it, detect it, punish it. That reaction is understandable, but it rests on a weak foundation. Detection tools produce false positives, which is unfair to honest students, and false negatives, which rewards the dishonest. They also struggle with non-native English writing and with Arabic, which matters in Gulf institutions where students write in both languages.
A more durable approach starts from a different question: not "how do we catch students using AI?" but "what do we actually want students to be able to do, and how do we know they can do it?" That question leads to assessment design, not surveillance.
AI literacy as a graduate attribute
Employers in Dubai, Riyadh, and beyond now expect graduates to use AI tools competently and responsibly. A university that forbids those tools in every course sends students into the workplace unprepared.
AI literacy means more than knowing how to write a prompt. It includes understanding how models produce text, why they fabricate references, how bias enters outputs, when use is appropriate, and how to disclose it. It also includes the discipline-specific judgment to evaluate an AI-generated answer in law, medicine, engineering, or Islamic studies, where a confident-sounding error can have real consequences.
The practical step is to treat AI literacy like academic writing: taught early, reinforced in every faculty, and assessed.
Redesigning assignments
The take-home essay on a general topic is the assignment most exposed to AI. Instead of abandoning essays, instructors can make them harder to outsource and more valuable to complete. Oral defences, in which a student explains and justifies their written work, test understanding directly. In-class writing under supervision remains useful for core skills. Assignments tied to local data, a specific case, a field visit, or a class discussion cannot be answered by a model that was not there.
Some instructors now ask students to use AI openly: generate a draft, then critique it, correct its errors, and submit both with a reflection. This turns the tool into an object of study and makes the student's own thinking visible.
Whatever the design, the rule should be stated per assignment: AI not permitted, permitted with disclosure, or required. Ambiguity is where misconduct cases are born.
AI as a research assistant
For faculty and graduate students, AI can accelerate literature searches, summarize long documents, suggest code, clean data, and improve the readability of a draft. These are real gains, especially for researchers writing in a second language.
The limits are equally real. Models invent citations, misstate findings, and cannot judge whether a method is sound. Every reference must be checked against the original source, every claim verified, and every use disclosed where the journal or the university requires it. Confidential data, including unpublished results and personal information from research participants, should never be pasted into public tools.
Institutions such as the Mohamed bin Zayed University of Artificial Intelligence in Abu Dhabi and the research universities of Saudi Arabia are already producing guidance along these lines; smaller institutions can adapt it rather than start from scratch.
Writing a policy that people actually follow
Good policy is short, specific, and consistent across departments. It defines permitted and prohibited uses, requires disclosure, explains how suspected misuse is handled, and protects students from accusations based on a detector score alone. It is reviewed every year, because the tools change faster than academic calendars.
Crucially, policy should be written with students, not only for them. Students who help draft the rules understand them, and are more likely to defend them among their peers.
How a university can start
A full transformation is not required in the first semester. A workable sequence looks like this:
- Form a small working group of faculty, students, IT, and the registrar, with a clear mandate and a deadline.
- Publish a one-page interim policy and a per-assignment disclosure template within one term.
- Pilot redesigned assessments in five to ten courses across different faculties, and collect feedback from instructors and students.
- Offer a short AI-literacy module to all first-year students, in Arabic and English where relevant.
- Review the results, revise the policy, and scale what worked.
The universities that will thrive are not those that keep AI out, but those that teach students to use it with honesty, skill, and judgment.
