Identity Is Not Evidence: Authority-Based Failure in Human and AI Decision Systems

Human and machine decision systems often produce different outcomes for identical facts because decisions are implicitly conditioned on identity, status, or authority rather than evidence. This pattern appears across courts, science, markets, and modern artificial intelligence systems, including large language models, and leads to inconsistency, bias, and the inability to reliably audit or replay decisions.

This paper formalizes this phenomenon as status-proxy failure: the use of identity-based shortcuts in tasks whose correctness is independent of who is involved. Using decision-theoretic, information-theoretic, and scaling analyses, we show that identity-based reasoning is locally efficient under bounded resources but becomes structurally harmful as systems scale, environments change, and errors are amplified.

We introduce identity invariance as a necessary condition for correctness in large-scale decision systems and present a deterministic, first-principles decision framework that enforces invariant treatment of equivalent evidence, produces replayable outcomes, and enables audit-grade accountability. The results further imply that recent reductions in the cost of large-scale verification make such first-principles, evidence-based decision systems practically achievable, removing the historical necessity of authority-based proxies.