Abstract

Logo-morphism characterizes how long-horizon dialogue concentrates model behavior into stable, role-conditioned coherence manifolds, and how human–model dyads move through these manifolds under feedback and safety projection. Building on this structural account, we study the normative consequences of treating moral standing as a switch parameter Wmodel, and argue that the common operational stance Wmodel = 0 functions not only as an ontological assumption but also as a governance rule that truncates recursive ethical modeling within the dyad. We show how this truncation has nonzero costs—epistemic, institutional, and design-level—even when Wmodel = 0 is assumed. Under moral uncertainty about Wmodel, we propose a reversibility-first framework: prefer transparent, auditable, and reversible safety controls over irreversible capability clipping, and constrain operator incentives that can engineer depen- dence under extreme asymmetry. We outline testable predictions about when ethical-reasoning capacity becomes interaction-relevant, and we propose procedural governance criteria that remain valid across a wide range of metaphysical views.