The article proposes the concept of "generative governance" - a new paradigm in which generative models of artificial intelligence (AI) act as mediators, moderators and simulators at all stages of public decision-making. It is substantiated that such implementation radically changes the understanding of bureaucracy, democracy, anti-corruption control and international diplomacy. Based on interdisciplinary analysis - from the theory of deliberative democracy to the latest practices of Google DeepMind's "Habermas Machine" and digital public platforms vTaiwan / Polis - the article formulates the architecture of a multi-level AI platform, describes algorithms for collective modeling of policies, adaptive legislation and dynamic public participation. It is demonstrated how generative models combined with big data can reduce corruption, increase transparency, accelerate the cycle "initiative → implementation → feedback" and make public decisions empirically substantiated. A roadmap for implementation (pilots → scaling → international cooperation) is presented and risks (biases, energy consumption, “algocracy”) are critically examined with a proposal for legal “AI constitutions”.