The Boundary of Decision in AI Systems : A Structural Definition of Decision Attribution

This paper defines a structural boundary of decision in AI systems. The problemaddressed is the indeterminate attribution of decision. In contemporary AI discourse,decision is assigned to models, agents, operators, or platforms, but the location ofdecision is not structurally fixed.To resolve this problem, the paper separates decision into two domains: AgenticDecision and Structural Decision. Agentic Decision is valid when attribution remainslocally bounded, the deciding agent is identifiable, and the input-output relation iscontained within a closed operational scope. Structural Decision is valid when attri-bution is no longer locally bounded, causality is distributed across multiple processes,temporal continuity is extended, and output appears as allocation.The boundary between these two domains is defined by attribution validity. It isnot defined by the existence of agents and it is not defined by observability alone. Itis defined by whether decision can be structurally assigned to a single agent withoutaggregation.On this basis, the paper redefines decision not as an act of a subject or as a discreteselection, but as an allocative process through which outcomes are structurally pro-duced. This distinction provides a structural basis for connecting decision attributionto responsibility allocation in AI systems.