It is a comfortable story that students outsource work because they cannot be bothered. It is comfortable because it puts the problem entirely inside the student, where an instructor has no obligations.

The surveys do not support it. When students are asked in conditions where they have no reason to lie, the reasons cluster into three, and only one of them is about effort.

Time compression. Four assessments due in one week, a job, a commute. The model is not replacing the work of thinking. It is replacing sleep. That is a timetabling problem wearing an integrity costume.

Fear of the blank page. A surprising number of students describe using a model to get started rather than to finish, then not knowing where the line is once they have a draft in front of them that is better than what they would have written. The policy usually does not tell them, which we wrote about in the syllabus policy piece.

The task invites it. Some assignments are, honestly, a test of whether you can produce five hundred words of prose in a given shape. If the assignment can be completed without understanding anything, a tool that produces prose will complete it. That is not a moral failure on the student's part so much as a design fact.

The uncomfortable implication is that a meaningful share of AI use is a signal about the assessment rather than the student. The assignments that survive contact with generative models are the ones where the product was never the point: where you have to defend a choice, explain a method, or account for a decision you made in week three.

That does not make integrity rules unnecessary. It does mean that a policy without an assessment redesign is asking students to voluntarily do a harder version of a task the system does not require them to do. Some will. Most will not, and the ones who do will be quietly disadvantaged against the ones who do not.

The fix people reach for first is surveillance. The fix that actually holds is making understanding the thing you grade.