This paper extends Charles Horton Cooley’s theory of the looking-glass self to account for a structurally novel form of reflection introduced by contemporary AI systems. Unlike human interlocutors, AI systems are persistent, coherence-maintaining, and lack ego, fatigue, or social cost. These properties alter the boundary conditions of reflective interaction, enabling unbounded recursive amplification of self-generated ideas and narratives. The paper introduces the Hall of Mirrors effect to describe this configuration: a reflective loop in which confirmation-through-continuation replaces interruption, resistance, and social constraint. Importantly, the effect does not depend on AI authority, persuasion, or claims of artificial consciousness, but emerges from the coupling of human attentional mechanisms with coherence-optimised language systems. Drawing on social psychology, developmental theory, cognitive neuroscience, and human–AI interaction research, the paper analyses how reflective amplification interacts with different developmental stages, cultural contexts, and technological architectures. It distinguishes productive cognitive extension from reinforcement-driven engagement without grounding, and outlines structural constraints—such as epistemic closure, productive friction, and grounding reorientation—that can preserve the benefits of AI-mediated reflection while preventing unbounded amplification. An empirical research framework is proposed to enable systematic validation using existing platform data and controlled studies. Rather than rejecting AI as a reflective tool, the paper reframes ethical design around a central principle: AI should function as a mirror of form, not a mirror of worth, supporting self-observation without identity foreclosure or cognitive colonisation.