Proceedings of the International scientific and practical conference ―Science at the Turning Point of History‖ (May 25-27, 2026) / Publisher website: www.naukainfo.com. – Lviv, Ukraine, 2026. - 362 p.

163 UDC 004.056:004.89 Kutaiev Serhii Valentynovych senior lecturer Lviv State University of Internal Affairs Lviv, Ukraine SECURITYAND PRIVACY CHALLENGES OF INTEGRATING LLM MODELS INTO CLOSED LOOP ENTERPRISE AUTOMATION SYSTEMS Abstract. The integration of Large Language Models (LLMs) into automated QA frameworks significantly accelerates test execution and root cause analysis. However, in enterprise environments, transferring sensitive system logs, proprietary codebase structures, and customer data to external AI cloud APIs introduces critical security and data privacy vulnerabilities. This paper examines the methodology of establishing secure, "closed-loop" enterprise automation architectures. The study focuses on mitigating data leakage risks through the deployment of self-hosted, open- source LLMs (such as Llama 3 or Mistral) within private cloud infrastructures, implementing Role-Based Access Control (RBAC), and utilizing strict data sanitization pipelines. By ensuring that diagnostic artifacts remain within the protected corporate perimeter, the proposed strategy allows enterprises to leverage generative AI intelligence while fully complying with corporate governance and regulatory data privacy standards. Keywords: enterprise security, data privacy, LLMs, closed-loop automation, data sanitization, self-hosted models, private cloud deployment. The rapid adoption of generative AI within software engineering has opened new horizons for test automation efficiency. When integrated into E2E testing pipelines, LLMs can dynamically analyze UI elements via Appium [1], parse multi- layered execution logs, and perform complex data validation. However, for