Why AI Detection Fails for Academic Integrity
Jonathan A. Karr, Grigorii Khvatskii, Ting Hua, Nitesh V. Chawla · arXiv (2026) · doi:10.48550/arxiv.2608.11256 · gemini-3.1-flash-lite
TL;DR
Commercial tools designed to catch AI cheating often mislabel legitimate student writing that uses AI for basic grammar or clarity. In contrast, students who use AI to generate entire papers can easily bypass these detectors using readily available 'humanizing' software. This creates a trap where students doing the right thing are unfairly flagged, while those intentionally cheating remain undetected.
Method and sample: Dataset of 642 English research abstracts (2013-2015 vs. 2023-2025) sampled from OpenAlex across chemistry, computer science, political science, and theology. Tested Pangram 3.2 and GPTZero detectors, along with a GPT-5 Nano baseline, across three generative rewrite conditions and Undetectable AI v11 humanization.
#academic integrity#AI detectors#assessment validity#humanization#psychometrics#higher education