In this work, we aim to provide a framework that enables the operator of a cyber-physical system to assess its operation in the presence of data-driven, learning components. Towards this, we leverage Architecture Analysis & Design Language (AADL) to provide a representation of the system’s components and their interconnections to construct a modular environment allowing for the inclusion of different detection and learning mechanisms, which supports a full model-based development including system specification, analysis, system tuning, integration, and upgrade over the lifecycle. Specifically, we illustrate the propagation of errors throughout system components due to adversarial attacks in the learning processes for a UAV system and obtain safety tolerance thresholds.