Every major university in the English-speaking world now has an AI policy. Some are thoughtful. Many are not. In researching this analysis, we reviewed published AI detection and academic integrity policies from fifteen universities across the United States, United Kingdom, European Union, and Australia. What we found is a landscape of remarkable inconsistency: institutions facing the same challenges, arriving at starkly different solutions, and - in too many cases - adopting policies that prioritize institutional convenience over student rights.
The Detection Tool Landscape
The first question every university faces is which detection tool to use - or whether to use one at all. Of the fifteen institutions we analyzed, twelve mandate some form of AI detection for submitted student work. The remaining three - including one Russell Group university and one Ivy League institution - have explicitly declined to adopt detection tools, citing concerns about accuracy and equity.
Among those that mandate detection, Turnitin's AI detection feature is the most widely adopted, used at nine of the twelve institutions. GPTZero is used at two, typically as a secondary check. One university developed its own in-house tool, which it describes as "experimental."
Appeal Processes: The Critical Gap
The most significant variation - and the most consequential for students - lies in appeal processes. We found three broad categories:
Robust appeals: Three institutions have developed multi-stage appeal processes that include the right to a hearing, the right to present evidence of authorship, and the right to have the case reviewed by someone other than the original instructor. These processes typically resolve within two to four weeks.
Minimal appeals: Six institutions offer appeals, but the process is informal: the student meets with the instructor or a department representative, presents their case, and receives a decision. There is no formal hearing, no independent review, and no guaranteed timeline.
No clear process: Three institutions have AI detection policies but no published appeal process specific to AI accusations. Students at these institutions are directed to general academic misconduct procedures, which were designed for traditional plagiarism cases and often assume guilt rather than innocence.
Penalties: A Spectrum of Severity
Penalties for AI-detected submissions range from a required rewrite (at the mildest institutions) to expulsion (at the most severe). The median penalty is a zero on the assignment plus a formal notation in the student's file. Three institutions impose automatic suspension for a second offense, regardless of whether the first accusation was successfully appealed.
What Good Policy Looks Like
The best policies we reviewed share several characteristics: they treat detection results as a starting point for inquiry, not as evidence; they provide clear, multi-stage appeal processes with guaranteed timelines; they require human review before any penalty is imposed; they acknowledge the known limitations of detection tools in their policy language; and they publish data on detection rates and false positive rates.
No institution we reviewed gets everything right. But the gap between the best and worst policies is enormous - and it is students, not institutions, who pay the price for that gap.
Recommendations
Based on our analysis, we recommend that every institution adopting AI detection tools also adopt: a published appeal process with a guaranteed resolution timeline of no more than four weeks; mandatory human review before any penalty is imposed; annual publication of detection and appeal statistics; bias auditing of detection outcomes disaggregated by student demographics; and explicit acknowledgment in all policy documents that current detection tools produce false positives.
The technology of AI detection will continue to improve. But technology alone cannot solve what is fundamentally a policy problem: the question of how institutions balance integrity with fairness, and whether they treat their students as partners in learning or as suspects in an investigation.