Make AI rules concrete and versioned
A useful rule names permitted, restricted and prohibited practices in relation to the learning outcome. It also explains what students must record and which version applies. Broad bans or permissions leave common cases unresolved. Educators need components, examples and a visible change path.
Make independent work visible during the process
Independent work is not confined to the final document. Decisions, intermediate states, subject conversations and traceable contributions can form a more informative record. Evidence should not become continuous surveillance. It needs a defined learning purpose, advance notice, proportional data collection and a clear assessment function.
Review technical findings before academic judgement
A similarity report or AI score is not a completed conclusion. Educators must distinguish file, run conditions, type of finding and disciplinary relevance. Fair communication states what is known, what remains uncertain and which further information will be considered. False positives need a defined review and correction path.
Govern procurement, privacy and review
Before procurement, define purpose, data flow, contracts, deletion, reference databases and independent evaluation. A product demonstration is not a representative benchmark. Two-person review is likewise an adaptable control rather than a universal duty. Ombuds responsibilities must come from the institution’s actual structures.
Teach prevention and reproducibility as disciplinary work
Prevention belongs inside subject practice: students exercise source decisions, tasks require reasoned intermediate work, and reproducibility is discussed with appropriate criteria. AI-resilient does not mean making all tool use technically impossible. It means assessing the learning outcome through decisions and disciplinary explanation that can be examined.