Anti-Cyberbullying AI Chatbots: A Systematic Review of Criminological Theories, Evidence, and Criminal Justice Applications

Abstract

This systematic review synthesizes evidence on AI chatbots for cyberbullying prevention and intervention from a criminological perspective, with explicit attention to Asia Pacific applicability. Following PRISMA 2020 guidelines, we identified 25 studies (2018-2025) across seven databases and appraised each using criteria adapted from MMAT and JBI tools, assigning every source an evidentiary weight (High, Moderate, Low, or Indicative proof-of-concept). Four chatbot roles emerged: real-time detection, victim support, preventive education, and bystander intervention training. Disaggregated analysis tempers headline performance claims. The frequently cited 89-99% accuracy range reflects laboratory classification on benchmark datasets, predominantly English-language Western corpora, evaluated under balanced class distributions and within-dataset conditions. No included study reported AUC or class-imbalance-adjusted metrics, employed temporal validation, or evaluated systems in real-world deployment with adolescents. Where precision and recall were disaggregated, asymmetric error profiles emerged with criminologically distinct social costs that single-metric reporting obscures. Only three studies qualified as High evidentiary weight, all addressing user-centred design in Western individualist contexts; victim support claims rest predominantly on Indicative proof-of-concept demonstrations. Geographic and linguistic concentration is a defining characteristic of the evidence base: zero studies have validated systems for Southeast Asian languages, no participatory design has engaged Asian youth, and no work has evaluated Asia Pacific platforms (WeChat, LINE, KakaoTalk). Drawing on Asian and Southern criminology, we argue that the uncritical transfer of Western-validated systems risks ineffectiveness, discriminatory false positives, and failure to detect culture-specific harms. The paper offers five criminological practice implications, each framed as theoretically informed extrapolations requiring local validation. AI chatbots are positioned as potentially valuable complementary tools within human-led prevention ecosystems, conditional on rigorous local research, transparent metric reporting, and sustained interdisciplinary collaboration.