When students use AI to generate answers, the issue is incentive design, not dishonesty.
If assignments reward polish over reasoning, students simply responds to the signal we send.