Abstract: The rise of generative artificial intelligence marks the further progress of human society towards the era of strong artificial intelligence, which leads to not only the expectation of the "singularity", but also the hidden worry that technology governance will fall into the "Collingridge's dilemma". The risks of generative artificial intelligence technology can be staged into three layers: "initial-result-derivative", and they have dynamic characteristics, which can be specifically deconstructed into complexity, uncertainty and high liquidity. In the face of generative artificial intelligence dynamic risks, the traditional governance system faces difficulties in governance thinking, governance tools and governance system. And for the coordination goal of technological development and risk regulation, the concept of adaptive governance has become into being. In this way, it can not only break through the traditional governance dilemma and flexibly govern generative artificial intelligence dynamic risks, but also avoid being too aggressive and destroying the stability of the governance system. Specifically, the adaptive governance system of generative artificial intelligence dynamic risks includes the following contents: in terms of governance thinking, it emphasizes the distribution justice of dynamic risks; in terms of governance tools, institutional ethics and ecological technology go hand in hand; and in terms of governance system, compliance, empowerment and supervision are used as the main governance mechanisms for the risks at different stages, and a multi-subject hierarchical action network is constructed.