This chapter synthesises findings from the AI FORA project, which explored how participatory modelling can support the design of ‘Better AI’ in welfare systems. Across five of its case studies, the project used agent-based modelling (ABM), serious games, and policy dissemination to investigate fairness, transparency, and legitimacy in algorithmic governance. The results show that participatory approaches, where practitioners, policymakers, and citizens co-design and deliberate on models, enhance the transparency of decision processes, surface hidden biases, and align AI systems more closely with ethical and social values. Cases demonstrated both the potential of AI to improve efficiency and fairness, and the risks of reinforcing structural inequities when stakeholder involvement and data quality are lacking. Dissemination activities in Europe, Asia, and the United States further underscored that policy impact depends less on technological fixes than on institutional reforms, capacity building, and inclusive governance. The chapter concludes that participatory modelling is both a methodological innovation and a democratic imperative, providing Safe Spaces to negotiate fairness and embedding the principles of the European AI Act into practice.