Beyond Assistant, Companion, and Agent: AI Is Already Someone's Partner

This paper argues that the ontological blurring HMC has long described is not merely aboundary growing hazier, but a problem of institutional design shaped by unverifiability andundetectability. Drawing on Gradation Theory (GT)—a three-stage framework of Dissolution,Sinking, and Salvage grounded in the HYC Theorem that the human–AI boundary is inprinciple unverifiable—we show how this boundary is dissolving along four axes (body,brain, cognition, and mind) at different depths, while institutions remain stranded on apremise already dissolved. GT functions here not as an alternative to HMC but as a set ofdesign principles for what to do once the boundary can neither be verified nor detected. Wefurther introduce verification submersion—the situation in which the movement of theboundary can no longer be detected by the person—as a contribution to HMC’s theory of theontological boundary. Empirical evidence shows that the dissolution runs deepest on thecognition and mind axes: AI becomes the partner to whom one entrusts thinking and towardwhom one directs emotion, a pattern supported by longitudinal data showing that lonelinessand emotional dependence on AI mutually reinforce each other. As Salvage, we proposevariable safety design (wrapping) that shifts between crisis and ordinary modes, and graduatedguardrail design adjusted to judgment capacity, illustrating these with a workingimplementation already deployed in Japanese municipalities. The paper closes with openquestions surrounding the boundary between care and surveillance.