University AI Policies After 3 Years – Are They Actually Working?
From an academic integrity coordinator perspective: we now have enough data to evaluate the first generation of Canadian university AI policies. The verdict is mixed.
Policies that worked: transparency-based frameworks that required disclosure and demonstrated understanding. These correlate with higher student satisfaction and lower formal integrity complaints.
Policies that didn’t work: blanket bans without enforcement mechanisms. These correlate with higher informal (undisclosed) AI use, lower disclosure rates, and no reduction in actual AI use.
Detection-based enforcement: the data is consistently poor. institutions that relied on detection as their primary enforcement mechanism have higher rates of contested integrity findings, more formal appeals, and more administrative burden than transparency-based institutions.
The surprise finding: policies co-developed with students perform significantly better across all metrics. This wasn’t obvious three years ago but it’s the clearest signal in our cross-institutional data now.
6 Replies
Join the discussion.
Log In to Reply
the institutions that have moved toward process-based assessment rather than detection-based enforcement are reporting fewer integrity incidents and higher student confidence in the fairness of evaluation. the case study evidence is there - it's the institutional will to redesign that's lacking.
CEGEP perspective: the co-development model is harder at the CEGEP transition level than at university but not impossible. worth trying.
three years of data is also three years of missed opportunities to build student AI literacy. every year we spent on detection-focused policy was a year we didn't spend on teaching students to use these tools thoughtfully and ethically. the opportunity cost of the wrong policy framework is real.
the institutions that got this right banned nothing and assessed differently from day one. the ones that banned everything are now quietly walking it back. we know which approach works.
The student co-development finding aligns with what Russ reported from his secondary school and with broader research on participatory governance. This is actionable for institutions still revising their policies - the process matters as much as the content. Sharing with my graduate faculty.
Mark's timeline of failure modes is accurate for most of the institutions I'm aware of. year one: ban everything. year two: carve-out policies that nobody enforces consistently. year three: quiet abandonment of enforcement. we're heading into year three and the honest conversation about what detection-based enforcement can realistically accomplish hasn't happened yet at most schools.