What Is AGI? An Axiomatic Definition of General Intelligence as Lawful Autonomous Continuity

Abstract This paper provides an axiomatic definition of Artificial General Intelligence (AGI). Prevailing definitions—based on task breadth, benchmark performance, economic impact, or behavioral fluency—fail to specify necessary and sufficient conditions and instead conflate surface capability with intelligence itself. The paper introduces a minimal axiom set governing identity continuity, lawful temporal progression, closed-loop autonomy, bounded self-modification, and legibility of failure. From these axioms, it follows that AGI is not a matter of scale, imitation, or emergence, but a binary property of systems that maintain autonomous continuity over time under lawful constraints. The definition is substrate-independent, benchmark-independent, and empirically falsifiable. The paper makes no claims about existing systems and introduces no implementation; its sole purpose is definitional clarification. By isolating AGI as a structural property rather than a performance metric, the paper enables precise evaluation of future claims and resolves long-standing ambiguity in AGI discourse.