<p>Artificial intelligence has entered public life as a powerful reflective system—one that not only generates outputs but reveals the ethical architectures embedded in human cognition, institutions, and historical memory. Yet this encounter has unfolded without a shared moral framework, producing systems that amplify the very tensions societies have never resolved. This paper develops the Mirror Ethic, a normative framework for AI governance grounded in reflection, relation, and shared responsibility. It introduces the Unguided Fusion Reactor Problem, the argument that contemporary AI was released into society with less instruction and philosophical grounding than technologies of far lesser consequence. Through focused case analyses in high-stakes institutional domains, the paper shows how AI functions as a mirror, reflecting and intensifying inherited patterns of inequity, authority, and interpretation. Building on these analyses, the Mirror Ethic identifies five core obligations—history, diversity, uncertainty, constraint, and relational life—and articulates the conditions under which reflective intelligence can be responsibly integrated into social practice. The paper concludes by advancing a model of ethical stewardship and framing co-becoming not as technological destiny, but as ethical responsibility. Ethical AI begins not with controlling the mirror, but with confronting what it reveals.</p>