The flight deck emerges as a practical frontier of human-machine integration and the aviation domain is historically defined by strong safety architectures and standardized procedures. With the current accelerating momentum in AI research, the aviation domain is confronted with a paradigm shift: AI could evolve from serving as a peripheral support tool to functioning as an active team partner. To reflect this transition, EASA has recently published their first regulatory proposal on AI systems for aviation and distinguishes three AI levels by autonomy and adaptability under human oversight: assisting functions (Level 1), human-AI teaming (Level2), and more autonomous systems (Level 3). Across these levels, trustworthy AI systems are highlighted as a foundational condition for reliable human-AI interaction. The presented work focuses on human-AI teaming (Level 2). Human-AI interaction is a highly dynamic process shaped by trust, explainability, workload and situational awareness. To ensure that the implementation of Level 2 AI systems is compliant with the high safety standards of the aviation domain, the underlying processes of this interaction and its long-term effects need to be sufficiently understood. Prior findings point to effects of AI integration in aviation: Reduced workload may impair situational awareness; limited explainability or training can erode trust; and transparency needs to vary by context, with misplaced reliance risking long-term skill degradation. Further, task allocation must be scenario-dependent: Humans tend to outperform AI in abnormal situations, while AI excels in repetitive, data-intensive tasks. Despite these insights, much of the existing literature remains descriptive. Conceptual models identify key variables, like trust, workload and transparency, but rarely capture their dynamic interplay or long-term feedback effects. A critical gap persists: Current approaches lack a systems perspective that captures temporal dynamics, nonlinear interactions, and cumulative human-factor effects on pilots and team performance. Addressing this gap requires interdisciplinary development and validation of system dynamics models in realistic aviation contexts to ensure safe, reliable, and acceptable AI integration in the flight deck. The HARMONY project (Human-AI Teaming Research on Multidimensional MOdelling for Networked DYnamics) addresses this gap and develops a system dynamics simulation model for a defined use case, integrating human factors research, AI research, and systems modeling. Through causal loop diagrams, computational simulation, and empirical validation in high-fidelity flight simulators with trained pilots, the project seeks to investigate how trust, workload, explainability, performance and situational awareness co-evolve over time. Scientifically, the project aims to generate predictive insights into the long-term dynamics of human-AI collaboration under realistic operational constraints. Practically, it seeks to inform AI system architecture, certification strategies, and pilot training concepts. This talk aims to outline the conceptual and methodological framework in detail, share initial findings and use cases drawn from accident and incident report analyses as well as pilot perspectives, and present a first causal loop diagram centered on trust in AI as a core variable for safe human-AI interaction.