Proceedings of the International scientific and practical conference ―Science in the Era of Globalization‖ (May 22-24, 2026) / Publisher website: www.naukainfo.com. - Zurich, Switzerland, 2026. - 353 p.

115 dynamic windows, and branching workflows, managing these individual scripts leads to severe code duplication and fragile test suites. Model-Based Testing (MBT) addresses this by abstracting the application‘s behavior into a formal model, typically a Finite State Machine (FSM) [2] or a directed graph, where states represent UI screens and transitions represent user actions. Despite the conceptual benefits of MBT, its widespread adoption has been hindered by the high cognitive load required to build and update these behavioral models manually. Generative AI effectively bridges this gap by automating the model extraction phase. Within a C# .NET testing infrastructure, the automation framework can programmatically dump the application's XML window hierarchy - the PageSource - at any given state. By feeding this structural data alongside visual cues into a generative Large Language Model (LLM), the AI can autonomously identify actionable elements (buttons, inputs, menus) and infer the relationships between different application states. The model is no longer drawn by hand; it is dynamically generated and updated by AI as it explores the desktop application [3]. Once the behavioral model is established by the generative AI, the framework transitions from model creation to intelligent path execution. Instead of running predefined test scripts, a central test engine uses graph traversal algorithms to generate execution paths on the fly based on specific testing goals, such as maximizing boundary coverage or targeting newly modified features. The AI agent translates these abstract paths into concrete C# code executions, interacting with the desktop client via Appium commands. If a modification in the desktop application introduces a new state or removes an old transition, the generative AI detects this variance during the next scan, updates the underlying model, and automatically adjusts the generated test scenarios without manual rewriting. Furthermore, integrating generative AI into the MBT pipeline significantly enhances the mutation and exploratory testing capabilities of the automation suite. The generative model can be prompted to deliberately introduce unexpected input sequences, simulating erratic user behaviors that a human QA engineer might not anticipate. This combined approach ensures that the desktop application is validated