AGDAU: Ontological Geometry of Admissible Action for AGI Systems

This publication introduces AGDAU (Ontological Growth-Only Admissibility) as a necessary condition for the safe existence of Artificial General Intelligence (AGI) systems. The core claim is that AGI safety cannot be guaranteed solely through policy alignment, reward design, or external control mechanisms.Instead, safety must be enforced at the ontological level by restricting the effective action space to growth-admissible actions only. Formally, AGDAU constrains the effective action space A_eff to a subset A' ⊂ A, defined by a growth functional G(s,a) such that only actions with G(s,a) > 0 are ontologically admissible. This publication serves as an anchor document.The definitions and formal structure introduced here are intended to remain stable and be referenced by subsequent specifications, benchmarks, compliance frameworks, and implementation layers. Attached files include:- A typeset PDF with formal notation.- A plain-text TXT version serving as a canonical, machine-readable reference. AGDAU is presented as a necessary, but not sufficient, condition for AGI safety.