Two decisions, and they're yours
Assessment design, expressed as two choices.
1. How much AI the educator has designed into the task.
You choose the AI Use Level from the five design levels of the Artificial Intelligence Assessment Scale (AIAS), so it reads like the policy your institution has already written. Based on the Artificial Intelligence Assessment Scale (AIAS), developed by Perkins, Furze, Roe and MacVaugh.
The range runs from work where no AI use is permitted, through work where it's allowed for planning or editing, to work where using it is the point of the exercise. Your discipline and your learning outcomes decide which applies.
2. Validity Threshold: what the work needs to reach
A weekly reflective post and a final-year dissertation shouldn't be held to the same bar. You set how much demonstration is enough for this piece of work.
Essential, the lightest setting. Suited to formative and lower-stakes work, where you want confirmation the work is the student's own without asking much of them.
Comprehensive, the middle setting. More evidence is required before a submission is validated. Suited to most assessed work.
Definitive, the strongest setting. The most evidence is required across all three dimensions. Suited to high-stakes summative assessment.
You choose based on the risk and learning objectives of the assignment.
Chosen per assessment, by the educator who set it.
What follows from your choices
Students see what's permitted before they start, which on its own resolves a good deal of the ambiguity they currently navigate by guessing. And the finding that comes back to you is expressed against the level you set, not against a generic standard.