Modern Paradigms of Application and Data Security: Information Protection Architecture Using Artificial Intelligence Algorithms

The purpose of the study is to substantiate a multi-level cyber defense architecture that combines signature rules, behavioral analysis, machine learning, and expert decision control. The study is aimed at determining the composition, functions, and relationships of components capable of protecting program code, data, training samples, models, and execution infrastructure. The methodological basis is the system, comparative, and structural-functional approaches. Scientific works are summarized in the areas of intrusion detection, behavioral analysis, adversarial attacks, model explainability, and practical implementation of intelligent tools. The components of the proposed architecture are compared with the stages of requirements formation, design, development, testing, deployment, and operation. Scientific novelty. A holistic protection model is substantiated, in which control of program components, telemetry, training sets, and models is carried out within a single process. A six-level structure is proposed, covering data collection, data preparation, intelligence, interpretation of results, response, management, and audit. The limits of automation are determined: reversible operations can be performed directly. At the same time, actions that can violate the availability or integrity of the system require the approval of a responsible specialist. Requirements for controlling the origin of training data, model versions, digital signatures, drift, and resistance to competitive influences are specified. Conclusions. It has been established that no single mechanism provides sufficient stability for a distributed software environment. Effective protection is achieved through the coordinated use of signature, adaptive, modular, and explainable tools. The recommendations provided include checking data sources, controlled retraining, testing model stability, logging decisions, and a phased increase in automation. The practical significance of the results lies in the possibility of their use when designing corporate, cloud, and critical systems. The proposed approach is also suitable for improving continuous integration, deployment, and incident response processes.