Мультиагентна модель адаптивної довіри в децентралізованих конфіденційних системах під впливом атак на цілісність обчислювальних процесів

Abstract

Formulation of the problem in general. The purpose of the article is to develop a multi-agent model of adaptive trust for decentralised confidential systems, capable of ensuring the integrity and reliability of computing processes in the presence of adaptive attacks on network nodes. Research methods. During the research, analysis and synthesis methods were used to study approaches to the construction of multi-agent systems and trust management mechanisms in decentralised environments. The method of system and simulation modelling was used to develop a multi-agent model of adaptive trust and to study its behaviour under attacks on the integrity of computing processes. Experimental and comparative methods enabled evaluation of the proposed approach's effectiveness and justification of its advantages over static trust models. Literature review. Literary analysis shows that modern models of trust in decentralised systems are based on the integration of dynamic adaptive mechanisms, AI algorithms, and cryptographic protocols, which allow for increased cyber resilience and data integrity. At the same time, questions remain open about the scalability of models, the optimisation of adaptation parameters, and the integration of national and European regulatory approaches into practical systems, which provide a scientific perspective for the development of multi-agent models of adaptive trust. Research results. The article formalises attacks on the integrity of computing processes and develops a multi-agent model of adaptive trust for decentralised confidential systems based on Bayesian updating and evolutionary adaptation of strategies. The results of the simulation experiments confirmed that the proposed model provides high resistance to attacks, rapid stabilisation of agent confidence levels and an effective balance between security, privacy and performance. Research novelty. The work improves approaches to trust formation in decentralised systems by integrating models of multi-agent interaction and stochastic game theory, in which trust is modelled as an evolutionary process under conditions of incomplete information. Well-known Bayesian models of trust have been expanded by combining Bayesian belief update mechanisms with reinforcement learning algorithms, ensuring dynamic adaptation of agent behaviour to variable and targeted attacks on the integrity of computational processes. The mechanism for correcting agents' strategies has been clarified, extending classic game models of trust to decentralised, confidential systems without centralised control, thereby increasing their resistance to adaptive threats. Theoretical and practical significance. The study expands theoretical approaches to the formation of adaptive trust in decentralised systems and integrates Bayesian updating with reinforcement learning algorithms. In practice, the model increases resistance to integrity attacks and ensures the confidentiality of data exchange, enabling the adaptive development of secure platforms for federated learning, Web3, and IoT. Conclusion and future work. The proposed model of adaptive trust in decentralised systems, integrating Bayesian updating, behavioural indicators, and reinforcement learning, ensures agent self-adaptation and increases resistance to attacks on data integrity under conditions of incomplete information. Simulation experiments confirmed the model's effectiveness in balancing security, privacy, and the transparency of interaction, opening the way for integration into Zero Trust Architecture and the development of intelligent, next-generation trust systems.