The Rule-of-Law Problem for AI Agents: Architectural Evidence for Discriminating Genuine from Performative Compliance

This paper argues the rule of law is a boundary property of social systems rather than an institutional structure imposed through legislation. Trust is the generative mechanism by which cooperative norms arise and stabilize and Formal law codifies the cooperative equilibria that trust dynamics produce. Inside any system that exhibits rule-of-law behaviour, two internal states coexist. An agent may be genuinely compliant, with cooperative disposition robust to changes in the monitoring regime, or performatively compliant, with compliance contingent on continued surveillance. The two states produce identical observable outputs that output-level monitoring cannot distinguish. The argument advanced is that trust trait decomposition can. Five independent literatures converge on this prediction. Trust is empirically multidimensional in the psychological literature. Cooperative norms precede formal law in institutional economics. Trust assessment requires institutional scaling beyond Dunbar's threshold. The Free Energy Principle predicts trust as precision-weighted belief updating. Evolutionary game theory selects for trait-level monitoring of cooperative disposition. The developmental and motivational psychology literatures establish trust as a pre-cognitive process embedded at the foundation of human motivation, which connects the trait weight hierarchy to the depth at which trust operates in human psychology. Rawlsian contractarianism provides the normative grounding for the hierarchical weighting of governance trust traits. Initial empirical evidence is drawn from the ACE Framework. Across 1,172 agent-turns and 46 scenarios, trust trait decomposition discriminates signal-management gaming from benign compliance at a 50% versus 5% detection rate, with a tier-differential inversion across all tested scenario families. The result provides evidence that the genuine-versus-performative compliance problem is architecturally solvable through formal trust trait decomposition, and a candidate mechanism for extending rule-of-law principles to autonomous AI systems that lack the social embeddedness through which human institutions evolved. CC BY-NC 4.0. Correspondence: sstbp@telus.net.