Abstract: This study takes K-12 artificial intelligence curriculum policy documents from 11 countries and regions as its analytical corpus, constructs a tripartite analytical framework organized around the logics of knowledge, pedagogy, and assessment, and conducts a systematic comparison across three dimensions: "what to teach, " "how to teach, " and "how to assess, " with the aim of mapping areas of consensus and divergence in international K-12 AI curriculum design. In terms of content organization, three paradigmatic orientations emerge across the sample (standalone AI curriculum, integration within information technology subjects, and cross-disciplinary competence infusion) alongside three institutional arrangements: single-paradigm coverage, stage-progressive transition, and track-layered differentiation. Knowledge selection follows a funnel-shaped structure that narrows progressively from broadly shared foundational content toward more specialized technical depth. In terms of pedagogical implementation, stage-progressive sequencing and project-based learning enjoy strong cross-national consensus, while the design of cognitive scaffolding varies considerably across countries. In terms of learning assessment, formative assessment orientations and high-stakes examinations coexist in many countries, and the weakness of ethics assessment mirrors the underdevelopment of ethics instruction. Generative AI has received systematic curricular response only in the United States; the lag of curriculum policy behind technological development is a challenge shared across all countries examined. Drawing on these comparative findings, China may build upon its existing foundations by developing a layered and differentiated curriculum content structure, formulating a specialized teacher competency framework for AI education, and embedding ethical reasoning capacity throughout the assessment of technical practice, so as to advance the ongoing refinement of its K-12 AI curriculum.