Work

Управление рисками искусственного интеллекта: концептуальная модель проактивной безопасности в условиях киберпреступности с использованием искусственного интеллекта

Year: 2026
Type: article
Abstract: The rapid integration of artificial intelligence technologies, including generative models, into critical business processes is accompanied by the growth of specific risks that cannot be effectively neutralized within traditional reactive risk management models. Existing approaches, including the National Institute of Standards and Technology Artificial Intelligence Risk Management Framework and the European Union Artificial Intelligence Act, are predominantly recommendatory or regulatory in nature and do not offer integrated mathematical models for proactive forecasting and quantitative threat assessment. Research objective: development of a conceptual model and framework for proactive artificial intelligence risk management that integrates artificial intelligence methods for cybersecurity to predict incidents and ensures unity of security, ethics, and regulatory compliance tasks. Methodology. The paper presents a systematic analysis of modern artificial intelligence risks (data-related, model-related, operational, ethical risks), their classification according to danger level and autonomy of the artificial intelligence agent. A formal statement of the risk management problem as an optimization problem is proposed. A three-level framework has been developed that combines predictive analytics methods, natural language processing for threat analysis, and deep learning for anomaly detection. Results. The authors propose an original classification of threats related to the use of artificial intelligence for criminal purposes. A risk management system model has been formalized. Key gaps in current legislation have been identified, particularly in the area of liability for decisions made by autonomous systems. Conclusions. The necessity of a paradigm shift from reactive to proactive strategies has been proven. It has been established that a hybrid approach to risk assessment increases efficiency by twenty-five to thirty percent according to the F1-Score metric. Practical recommendations for implementing the proposed framework and directions for further research in the field of explainable artificial intelligence and synthetic data generation have been developed.
Language: Russian
Open Access status: green
OpenAlex ID: W7170123965
FWCI:
Cites: 7
Cited by:
Related to: 9