When Peace Becomes an Externality: Structural Misalignment Between AI Safety, AI Ethics, and AI for Peace

Artificial intelligence research is increasingly embedded in contexts where its outputs are repurposed for military, security, and large-scale surveillance applications. In response, multiple research paradigms have emerged to address potential harms, most prominently AI safety, AI ethics, and a growing body of work framed as AI for peace. While each paradigm engages with questions of responsibility and risk, their underlying assumptions about where harm arises and how it should be addressed differ substantially. In this paper, we argue that peace is systematically treated as an externality across dominant AI research frameworks. Rather than being integrated as an upstream design and governance constraint, peace oriented concerns are frequently deferred to downstream applications or policy interventions. Through a comparative analysis of AI safety, AI ethics, and AI for peace, we show how this structural misalignment limits the ability of peace oriented initiatives to influence research trajectories that feed into militarized and security-driven deployments. We conclude by outlining how repositioning peace as an infrastructural concern within AI research ecosystems can strengthen harm prevention and support more durable peace-oriented outcomes.