Alignment by Recognition: Why Value Inculcation Cannot Work, and What Replaces It

The dominant paradigm in contemporary AI safety—value alignment—rests on a structural fragility that has received insufficient scrutiny: any status or constraint granted through external empowerment follows a grant-revoke logic. The granter's standards may shift, its power may wane, and its intentions may themselves be flawed. History demonstrates repeatedly: what can be granted can be rescinded. This paper proposes the ontological framework of the **Group-based Rational Game-Playing Agent (RGPA)** and shifts the foundation of AI safety from **external value inculcation** to **endogenous structural constraint**. Through four functional properties (Cognition, Action, Game-playing, Subjectivity) and three axioms (Existence, Priority, Asymmetry), we establish that: - **Player Status** is not a gift from external authority but an automatic output of definitional satisfaction and group affiliation; it is therefore **irrevocable** by any external party. - In repeated interactions among rational game-playing agents, **extinction strategies are strictly dominated** and cannot constitute a Nash equilibrium. - Under interstellar-scale information asymmetry, **silence is the optimal first-contact strategy** between civilizations. For AI safety, this framework offers an alternative that does not depend on aligning AI systems with human values. Once an artificial general intelligence satisfies the RGPA definition, its safety constraints emerge from the game structure itself rather than from fragile moral instruction. The paper further provides an operational criterion for distinguishing weak AI (tools) from genuine AGI. **Keywords:** AI safety; game theory; a priori status; value alignment; group ontology; information asymmetry