This paper introduces Relational Control as a unified execution governance framework for both Human–AI and AI–AI systems. Conventional Human–AI interaction systems and multi-agent AI systems have historically evolved as separate research domains, emphasizing autonomy preservation, usability, coordination, and stability as distinct objectives. Despite these differences, both domains share the same structural limitation:execution is assumed as the default response to input. The paper demonstrates that failures in both domains — including excessive intervention, over-execution, recursive feedback loops, interference, and autonomy loss — arise from the same underlying execution assumption problem. To address this limitation, the paper formalizes Relational Control as a unified pre-execution control framework in which execution decisions are conditionally governed through shared relational conditions across all interacting entities. The framework establishes:- relationally conditioned execution,- unified execution governance,- dynamic evaluation alignment,- relational history integration,- intervention regulation,- and non-execution as a first-class interaction outcome. Execution decisions are determined through relational state, relational dynamics, relational history, and dynamic evaluation criteria, producing outcomes including:- Execute- Delay- Delegate- Do Not Act Crucially, Human–AI and AI–AI systems are not fundamentally different interaction problems, but different instances of the same execution control structure. The proposed Relational Control Layer operates as a model-agnostic pre-execution control plane, decoupling execution decisions from inference processes across heterogeneous interaction systems. This publication serves as a foundational systems unification paper for the Relationship-Aware AI Research initiative, establishing Relational Control as the unified framework for conditional execution across Human–AI and AI–AI systems.
HARUKI ITO (Thu,) studied this question.