Detection is an arms race with a predictable ending. The more durable answer is assessment designed so that outsourcing it defeats the point.
For most of the history of higher education, a written artifact was reasonable evidence that a person understood something. That inference is now weak, and no amount of policy language restores it.
The policing instinct
The first institutional response was detection. It is understandable and mostly futile: detectors produce false positives on non-native writers, they degrade as models improve, and they place instructors in an adversarial posture with the students they are trying to teach.
A redesign instinct
The alternative is to change what we ask for. If the assessment requires a student to reason aloud, adapt to a follow-up question, or defend a choice against an alternative, then outsourcing it stops being useful — you cannot delegate the ability to think on your feet.
- Oral defense of written work, short and frequent rather than long and decisive.
- Assessments tied to a specific dataset, decision, or artifact the student produced.
- Process-visible work where reasoning is part of the deliverable.
- Explicit AI-use disclosure treated as method, not confession.
What I am doing about it
QuizBuddy exists because I wanted the oral option to scale past the arithmetic limit of one instructor and thirty students. It is still an experiment, and I report what does not work alongside what does.
Related