The Anchor Paradox: Stake, Steerability, and the Ontological Limits of Functionalist Accounts of AI Consciousness

Contemporary research on artificial consciousness is dominated by a functionalist indicator paradigm that overlooks an ontological foundation of subjective agency: stake. This paper advances the conditional thesis that if stake is a necessary condition for thick subjective consciousness, then programmable, reproducible, losslessly steerable computational systems cannot in principle satisfy it. Drawing on enactivism, non-equilibrium thermodynamics, and the phenomenology of life, it critiques Global Workspace Theory criteria line by line, formulates a steering counter-indicator thesis, and addresses the Swampman objection through a thin/thick consciousness distinction. The argument identifies a core ontological blind spot in functionalist frameworks: consciousness is not merely an architectural property, but a temporal achievement of ongoing worldly coupling. The relevant contrast is therefore not biological versus artificial substrate as such, but externally maintained versus endogenously self-producing modes of persistence. Artificial consciousness remains possible in principle, but only for systems whose self-maintenance, history, and world-involvement make them less reproducible and less losslessly steerable. This reframes the debate from functional resemblance to the existential organization that makes a continuing subject possible.Version 2 of the manuscript (Zenodo record version 3). Supersedes version 1; all changes are listed in Appendix A. Protocol, preregistration and registered results: https://github.com/simin-yuan/anchor-paradox