This is such a useful summary from Phil Dawson in this brief commentary in Medical Education:
The opening moves asked students not to use AI, to use it only in specified ways, or to declare how they had used it. Scales and traffic light systems proliferated. Validity was rarely the word used, since many in higher education treat it as a scary psychometric term best left to specialists,4 but it was still the underlying concept. If we know how a student used AI, we can make a better judgement about how well they have met the intended outcomes.
Others then asked whether an assessment system can be built on voluntary student compliance. Corbin and colleagues argued that “talk is cheap” and that what is needed is structural change: mechanisms that enforce the conditions of an assessment rather than just asking students to adhere to them. The University of Sydney5 pio- neered this through a two-lane model, in which educators may not restrict student use of AI unless the assessment is secured.
The conversation felt like it was over at that stage. It had the feel of the end of a dinner party, when the plates are stacked and people are patting their pockets for keys, and the remaining questions are ones nobody expects to settle tonight. What we had not accounted for was a guest who was never at the table, and who had no reason to honour anything agreed around it.
Wearable AI challenges the assumption that it is possible to sepa- rate students from AI at all. This is not a problem for the distant future: EssilorLuxottica sold more than seven million Meta AI glasses in 2025, over three times what it sold across the two preceding years combined. Hearing aids and earbuds are moving the same way. As these devices become more capable and harder to notice, validity models that depend on sterile, AI-proof moments of assessment become brittle. Inspecting every student’s eyewear and hearing aids is a threshold few are willing to cross, and signal blocking approaches are largely infeasible.
This is where the conversation is now. A question has been asked: How do we rebuild assessment validity for a time when students have uncontrolled and uncontrollable access to AI? A validity that does not depend on separability. There is a pause. Educators are looking around for someone who can answer it.
