From Linear Risk to Emergent Harm: Complexity as the Missing Core of AI Governance

This white paper examines why many contemporary AI governance frameworks fail despite good intentions. It argues that dominant risk-based approaches implicitly rely on linear models of causality and control, while AI systems operate as complex adaptive socio-technical systems characterised by feedback, adaptation, and emergent behaviour. As a result, governance interventions often do not eliminate harm but displace, transform, or conceal it across organisational, institutional, or temporal boundaries. Increased compliance may therefore coexist with persistent or amplified systemic risk. The paper proposes a complexity-based alternative to AI governance, centred on regulation as intervention rather than control, dynamic system mapping over static risk classifications, causal reasoning and simulation for policy design under uncertainty, and adaptive governance as institutional learning. Rather than offering prescriptive rules, the contribution provides a conceptual framework intended to support more robust, reflexive, and system-aware approaches to AI policy design and evaluation.