VER0: A Proposed Hybrid Framework for Reducing Bias Against Non-Native English Speakers in AI Text Detection

Artificial intelligence (AI) text detectors are increasingly used within educational institutions to identify AI-generated writing and uphold academic integrity. However, recent research has shown that many existing detection systems disproportionately misclassify authentic writing produced by non-native English speakers (NNES) as AI-generated. This paper proposes VER0, a hybrid AI text detection framework designed to reduce algorithmic bias while maintaining competitive detection performance. Rather than relying primarily on perplexity, VER0 combines multiple linguistic indicators—including calibrated perplexity, burstiness, repetitiveness (R-measure), reinforcement learning, and similarity scoring—within a supervised machine learning framework. The proposed methodology includes a comparative bias audit of existing commercial AI detectors, feature engineering across native and non-native English corpora, and evaluation using conventional classification metrics alongside a fairness index. By emphasizing equitable treatment of diverse writing styles, this work contributes to the growing discussion surrounding fairness, transparency, and responsible deployment of AI detection technologies in education.