AI governance in university assessment: from blanket bans to controlled use
A practical framework for deciding when AI should be restricted, permitted or declared in university assessment without replacing academic judgement.
Assessment policy is increasingly moving beyond a simple choice between banning AI and allowing unrestricted use. Different assessment designs require different boundaries.
A useful starting point is to define the purpose of the assessment. If the task is intended to measure unaided recall or an individual technical skill, external AI assistance may need to be restricted. If the task is intended to assess critique, synthesis or professional use of AI, bounded assistance may be appropriate.
The next step is to make the rule operational. Lecturers need a clear assessment-level setting, students need to know what is permitted before they begin, and any allowed assistance should remain within the intended boundary. Where declarations are required, they should become part of the assessment record rather than an informal afterthought.
Evidence should support review, not replace it. Writing history, session records, AI-use declarations, reference discrepancies and similarity signals can help an authorised lecturer understand how work developed. None of those signals, by themselves, establish misconduct.
Institutions should also decide what data is proportionate to collect, who can see it, how long it is retained and how students can raise concerns. These governance questions are as important as the technical controls.
A mature AI-assessment policy therefore combines assessment design, explicit boundaries, transparent student notice, proportionate evidence and human decision-making.