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.

164 enterprise-grade products - especially those in financial, healthcare, or government sectors - this integration poses severe security risks. Standard automated test suites running within C# .NET frameworks frequently process highly sensitive diagnostic data, including authentication tokens, proprietary API endpoints, intellectual property in the form of code logic, and personally identifiable information (PII). Transmitting these payloads to public cloud AI providers for processing creates unacceptable vectors for intellectual property theft and data breaches. To harness the capabilities of LLMs without compromising the security architecture, organizations must transition to a strict "closed-loop" automation paradigm [2]. The foundational element of this strategy is the replacement of public AI endpoints with localized or private-cloud deployments. By hosting open-source models (such as Llama 3 or Mistral) directly within secure enterprise infrastructures (e.g., Azure Private Endpoints or AWS VPC), the automated testing framework ensures that no data ever exits the corporate firewall. In this setup, the C# automation client communicates with the LLM instance via internally hosted, encrypted endpoints using standard TLS protocols, preventing external interception. The practical execution of this closed-loop strategy is realized through the deployment of AI infrastructure within dedicated On-Premises (On-Prem) data centers. By containerizing open-source language models using tools like Docker and orchestrating them via Kubernetes, enterprise infrastructure teams can provision local inference engines that operate entirely offline. This On-Prem architecture leverages enterprise GPU clusters (such as NVIDIA A100 or H100) and utilizes advanced optimization frameworks to ensure ultra-low latency during test run evaluations. Because the hardware, network layers, and model weights reside on physical, company-owned servers, the automation pipeline achieves complete immunity from external third-party service outages, unpredictable API throttling, and cross-tenant data contamination risks. Beyond infrastructure isolation, a robust enterprise security strategy requires the implementation of an intermediate Data Sanitization and Masking Layer. Before any execution artifact - be it an Appium stack trace, a network response payload from