Governing the Ungoverned: AI Safety Research as a Bridge Between Humanitarian Principles and Algorithmic Power

The rapid spread of artificial intelligence across military and humanitarian settings has brought into view a governance gap that is structural rather than incidental. AI systems are being built and deployed far faster than the legal, ethical and institutional frameworks needed to oversee them. This paper argues that AI safety research, understood as a policy and governance discipline rather than a technical specialism, offers a principled and practical way through this impasse. Drawing on the three thematic tensions identified by Humanitarian Alternatives, namely operational contexts, professional reconfigurations and interdisciplinary dilemmas, the paper shows that the frameworks and institutional proposals developed within the AI safety field address directly the challenges facing humanitarian actors. AI safety governance is not an adjacent concern. It is the connective tissue the humanitarian sector is missing, sitting between the principles it holds and the algorithmic systems it is increasingly being asked to rely on.