Risk-based AI regulation has become the dominant paradigm in AI governance, promising proportional controls aligned with anticipated harms. This paper argues that such frameworks often fail for structural reasons: they implicitly assume linear causality, stable system boundaries, and largely predictable responses to regulation. In practice, AI operates within complex adaptive socio-technical systems in which harm is frequently emergent, delayed, redistributed, and amplified through feedback loops and strategic adaptation by system actors. As a result, compliance can increase while harm is displaced or concealed rather than eliminated. We propose a complexity-based framework for AI governance that treats regulation as intervention rather than control, prioritises dynamic system mapping over static classifications, and integrates causal reasoning and simulation for policy design under uncertainty. The aim is not to eliminate uncertainty, but to enable robust system stewardship through monitoring, learning, and iterative revision of governance interventions.

White Paper / Policy Brief (Working Paper). Published version available at: https://doi.org/10.5281/zenodo.17929014