Defining AI Models and AI Systems: A Framework to Resolve the Boundary Problem

Emerging regulations assign distinct obligations to different parties along the AI value chain. For example, The EU AI Act differentiates between general-purpose AI (GPAI) models versus AI systems, and their providers or deployers have distinct obligations. Nevertheless, the term \textit{AI model} is itself undefined in the Act, even though it is used to define \textit{GPAI model}. This gap creates regulatory ambiguities, such as when and how an entity acquires provider obligations by modifying a model. Drawing on a systematic review of 896 papers and a manual review of regulatory documents across major jurisdictions (included in the Appendices), we propose conceptual and operational definitions of the terms \textit{AI model} and \textit{AI system} for contemporary neural network-based AI.