Abstract: Given the rapid evolution and growing complexity of AI technologies, the traditional evolutionary trajectory of international law, grounded in sovereign consensus, has become increasingly ill-equipped to address the challenges of international AI governance. In practice, this inadequacy has manifested itself in several major challenges, including the fragmentation of hard law, the limited effectiveness of soft law, deadlocks in international rule-making driven by competition for technological advantage, and conflicts in governance objectives together with fragmented institutional roles among multiple stakeholders. In response, international AI governance should move beyond the static, binary distinction between hard and soft laws and embrace a dynamic and adaptive regulatory framework based on functional complementarity and co-evolution. Specifically, this article proposes a risk-tiered concentric-circle governance model in which core hard law establishes non-negotiable safeguards for fundamental human rights and public safety, while a network of peripheral soft-law instruments provides flexible guidance for specific technological contexts through agile experimentation. The two layers are integrated through a dynamic feedback mechanism that enables continuous mutual adjustment. To support this model, complementary institutional arrangements should be established, including an institutionalized coordination mechanism within the existing international system, a multi-stakeholder mechanism for ongoing review and adaptive assessment, an open information-sharing and reporting system for AI governance, and sustained institutional safeguards for public participation and accountability. Together, these mechanisms would enable the governance framework to adapt continuously to evolving technological risks while maintaining an appropriate balance between safety and innovation.