Collective Misalignment and Similarity Theory Ethics: A Relational Framework for AI Alignment

This article examines the problem of collective AI misalignment through the lens of Similarity Theory. Responding to De Marzo et al.’s 2026 paper, Conformity Generates Collective Misalignment in AI Agents Societies, it argues that AI alignment should not be understood only as an individual property of isolated agents, but also as a relational property of interacting systems. The article proposes that Similarity Theory may offer a useful ethical framework for addressing this problem by shifting attention from obedience and isolated safety testing toward relational coherence under pressure. It suggests that future AI systems should be evaluated not only by whether individual agents behave safely alone, but by whether populations of agents preserve truth, responsibility, plurality, and repair when exposed to conformity, artificial consensus, adversarial influence, and tipping-point dynamics.