AI detection tools can incorrectly flag human-written essays as AI-generated, particularly when the writers are non-native English speakers, according to research led by Stanford’s James Zou. In the study described, more than 61% of 91 essays by students preparing for an English proficiency exam were classified as AI-generated on average by seven detectors. The findings raise concerns about relying on automated detection scores to determine academic misconduct, especially when students’ records may be affected.
