Systemic Vulnerability: From AI Systems to Environmental Systems

Abstract Complex systems – from deep-learning models to ecological networks – share an often-overlooked property: systemic vulnerability , an inherent susceptibility to cascades that stems from their very structure and interdependencies. While the normative literature typically frames this problem in the means of systemic risk , this article argues that risk is merely the symptom of a deeper, present condition. After mapping how “systemic risk” has migrated from economics into cybersecurity, environmental science and, most recently, the European Union’s AI Act , the paper reconstructs the philosophical genealogy of “system” in Kant, Hegel, cybernetics and complexity theory to show why opacity, feedback loops and emergent behaviour make complete causal mapping impossible. It then demonstrates, through the examples of frontier AI models and ecosystems, how small perturbations can propagate across dense relational networks, entangling technical failures with social inequities and environmental externalities. Because vulnerability is structural and ongoing, compliance-oriented risk controls remain insufficient. The article therefore calls for a reflexive ethic of vulnerability: governance practices that embed transparency, traceability, participatory oversight and sustainability metrics throughout the AI and environmental value chains. Recognising and managing systemic vulnerability, rather than attempting to eliminate it, is essential to safeguarding the concrete vulnerabilities of human beings.