Balancing Transparency, Cybersecurity, and Innovation: Technical and Legal Challenges of AI Regulation in the European Union

Abstract

This research presents an analysis of a fundamental problem in contemporary digital regulation, namely the need to ensure the transparency of algorithmic systems, maintain a high level of user cybersecurity, and foster technological innovation. The research focuses on the implementation of the EU Artificial Intelligence Regulation (Regulation (EU) 2024/1689), with particular emphasis on the legal and technical analysis of Articles 13, 50, and 78 of the EU AI Act. Special attention is devoted to the conflict between explainable artificial intelligence (XAI) methods and the vulnerability of models to various cyber threats, including adversarial attacks, model inversion, and data exfiltration. A critical gap is identified between the high-level legal requirements of the Regulation and the absence of harmonized technical standards for their implementation, thereby creating risks of legal uncertainty for developers. The article provides a detailed examination of promising technological solutions, including differential privacy (DP) and federated learning (FL), as means of achieving «compliance by design». Based on a synthesis of academic research conducted over the past five years, a roadmap is formulated for the transition to the concept of XAI 2.0, which integrates security, transparency, and innovative flexibility into a unified ecosystem of trust and the safe deployment of AI systems.