When chemistry undergraduate Lauren Jager at Idaho State University ran her PhD application personal statement through several AI detectors, every tool returned a result of almost 100 percent AI-generated. She had written every word herself. To pass the checks, she deliberately made her writing less polished, ultimately submitting a statement she considered inferior to her original. She was later accepted to a PhD programme at the University of Utah.
Her experience reflects a growing crisis in academic integrity technology. Universities deploying AI detection tools to police student submissions are doing so with instruments that researchers have found to be unreliable, biased, and easily circumvented. A 2025 study found that GPTZero, widely considered the most-used detector, produced a false-positive rate of around 16 percent on human-written essays. A 2023 study found that most tools performed inconsistently on human text and struggled more with output from advanced models like GPT-4 than older systems. Notably, the US Declaration of Independence has repeatedly been flagged as between 95 and 100 percent AI-generated.
Mike Perkins, who researches AI’s impact on academia at British University Vietnam, argued that even reasonably accurate detectors should not be used in high-stakes decisions given the false-positive risk. Marzena Karpinska of Simon Fraser University similarly cautioned that while detectors can identify broad trends across large datasets, they cannot reliably determine individual authorship.
A particular concern involves bias: a Stanford University study found that AI detectors incorrectly labelled more than half of essays written by non-native English speakers as AI-generated, with an average false-positive rate of 61.3 percent.
