Vietnamese High School Students in the AI Era: Examining the Control Threshold and Knowledge Paradox in Academic Integrity

Abstract As artificial intelligence blurs the boundaries between authorship and assistance, traditional frameworks for understanding academic dishonesty are becoming obsolete. This study employs a data-driven approach to decode the complex ethical decision-making of adolescents in a developing digital economy. Analyzing data from 923 secondary students in Vietnam, we combine the Theory of Planned Behavior (TPB) with machine learning (XGBoost) and SHAP interpretability techniques. The XGBoost model achieved superior predictive accuracy ( R 2 = 0.634) compared to traditional linear regression ( R 2 = 0.604). The study offers two major contributions to the discourse on technology and ethics. First, it exposes a ‘knowledge paradox’ where increased awareness of academic integrity rules is significantly associated with higher plagiarism frequency ( B = 1.12, p < .001), indicating a disconnect between digital literacy and ethical practice. Second, XAI analysis reveals a nonlinear threshold in Perceived Behavioral Control: when the perceived difficulty of avoiding plagiarism exceeds a critical tipping point (score > 3.5 on a 5-point scale), the likelihood of offending escalates disproportionately, regardless of students' moral attitudes. These findings suggest that purely informational campaigns are insufficient in the AI era. Instead, promoting integrity requires systemic interventions that address the psychological thresholds of control and the unintended consequences of digital literacy education.