Autonomy–Opacity Stratification: A Falsifiable Theory of Performance-Driven Functional Opacity in Advanced AI Systems

This paper argues that functional opacity in advanced AI systems can arise as a by-product of performance-driven delegation, before any deception or monitor evasion. H1 predicts a complexity-dependent crossover in the value of detailed intervention and a persistent opaque delegated-autonomy regime. H2 predicts higher downstream incidence of endogenous evaluation, selection pressure, and structural differentiation than in matched controls. H3 predicts a conditional order from opacity to evaluation, selection, and structural differentiation, with observer-relative selective visibility last and only under sufficient observation reactivity. Human admission timing, local closure, monitoring-budget reallocation, and epistemic depth are treated as modular derivatives. The governance target is managed rather than maximal transparency: semantic opacity is tolerated only while authority, boundary, resource, effect, residual, and revocation information remain verifiable. The paper presents a falsifiable theory and a staged test design; no experiment confirming the full chain is reported. The hypothesis is isolated from the author's broader Information Ecosystem Theory, which remains unpublished at the time of this release; this paper is written to stand alone.