Application of Artificial Intelligence in Identifying the Dangerous State of Offenders Using the HCR-20 Instrument: CBT-Based Treatment Strategies

Abstract

Introduction and Objective: The present study aims to develop an integrated conceptual model for applying artificial intelligence (AI) to the interpretation of HCR-20 indicators in order to identify and explain dangerous states among offenders and to design cognitive behavioral therapy (CBT)-based treatment strategies within Iran’s criminal justice system.

Methodology: In terms of purpose, this study is basic-applied, and in terms of nature, it is descriptive-analytical. It was conducted through a documentary approach and a systematic review of scientific sources. Data were collected by extracting information from relevant scientific articles, specialized books, theses, and domestic and international legal sources concerning four principal domains: HCR-20, artificial intelligence, CBT, and dangerous states. The data were analyzed qualitatively through qualitative content analysis, comparative analysis, and inferential analysis. Finally, an integrated conceptual model was developed on the basis of the findings of the systematic review.

Findings: The systematic review showed that each of the four principal domains has specific capacities and limitations and that none, independently, can comprehensively address the challenges of identifying dangerous states and designing effective rehabilitative interventions within the criminal justice system. HCR-20, as one of the established violence-risk assessment instruments, has acceptable validity and reliability but faces limitations such as low positive predictive value, limited sensitivity to clinical change, and a lack of randomized controlled trial evidence. AI can increase risk-prediction accuracy by processing large volumes of data and identifying hidden patterns, but it also raises serious concerns regarding algorithmic bias, opacity, accountability, and privacy. CBT, as one of the effective therapeutic approaches, has a positive effect on reducing recidivism, although its use involves challenges such as attenuation of effects over time, compulsory participation, and the gap between assessment and treatment. In Iran’s criminal justice system, the absence of a clear legal definition, the lack of a standardized risk-assessment system, weaknesses in correctional and rehabilitation programs, and an emphasis on punitive approaches were identified as major obstacles.

On this basis, the integrated AI-HCR-CBT conceptual model was developed around four principal components—input data, processing and analysis, analytical output, and therapeutic interventions—together with a feedback loop explaining the relationship among AI, HCR-20 indicators, and CBT. With characteristics including personalization, data-driven decision-making, dynamism, transparency, ethical orientation, integration, and adaptability to local conditions, the model can help address existing theoretical and practical gaps and enhance the accuracy, legitimacy, and efficiency of criminal-justice decision-making.

Conclusion: By proposing a comprehensive integrated model, this study takes a step toward evidence-based criminal policy and smarter criminal justice in Iran. The AI-HCR-CBT model can establish a coherent chain consisting of risk assessment through HCR-20, data-driven analysis through AI, and personalized therapeutic interventions through CBT, thereby potentially contributing to reduced recidivism, enhanced public safety, and fairer criminal justice. Implementation, however, requires localization, empirical validation, and appropriate legal, technical, and organizational infrastructure in Iran’s criminal justice system. Future studies should empirically validate the model in Iran, examine its application to specific populations, and develop interpretable AI tools for risk assessment.