Abstract: To understand the dynamics of artificial intelligence (AI) regulation in the United States and explore policy arrangements in the context of Sino-U.S. AI competition is crucial for China's participation in global AI governance and its pursuit of global technological leadership. However, existing research has predominantly focused on macro-level comparative analyses of national AI policies, with limited exploration into state-level AI policy comparisons. The U.S. regulatory policy framework, shaped by its federal political system, exhibits significant complexity and fragmentation. This unique structure offers valuable insights for policymakers in other countries navigating the complex and rapidly changing artificial intelligence regulatory policy development settings. Using Latent Dirichlet Allocation (LDA), Dynamic Topic Modeling (DTM), and content analysis, this study conducts an in-depth examination of 397 policy texts from 36 U.S. states, illustrating state-level regulatory preferences, policy impact areas, and shifts in policy focus over time. Along with the aforementioned, the study also compares the regulatory proximity of China's artificial intelligence policy to that of the United States at the state level. The results show that the U.S. has established a foundational AI regulatory policy system, characterized by seven policy topics and four core policy instruments. The seven policy topics include consumer privacy protection, public opinion and election, fintech regulation, public-private data governance, advisory committee building, automated decision-making fairness and system security and effectiveness. Among these, system security and effectiveness emerge as the most aligned topic between Chinese and U.S. AI regulatory frameworks. Beyond the seven policy topics, four key policy tools were adopted by states in the U.S., including checklist documentation, mandatory disclosure requirements, prohibition and risk classification. Artificial intelligence regulatory committees in U.S. states fall into two categories: comprehensive committees, like those in Oregon, Maryland, and Indiana, addressing broad AI issues; and specialized committees, such as New Mexico’s group on automated decision-making and New York’s panel on AI, robotics, and automation. The study also identifies that states such as California, Illinois, and New York exhibit the most diverse distribution of policy topics. In terms of regulatory proximity to China, Hawaii, Oklahoma, and New York rank highest, with technologically advanced states like California and Massachusetts also featuring prominently. Finally, drawing on the complexity of the artificial intelligence regulatory policy system in the state of the U.S. and the need for artificial intelligence development and regulation, it is recommended that China adopt policy innovation approaches that are tailored to regulatory scenarios and enrich the scenario hierarchy of the regulatory system, deconstruct the entire system into components and improve the details of the subject of regulatory policy, prioritize compatibility and symbiosis to align the value of policy tools, and emphasize collaborative efforts to enhance agile effectiveness in artificial intelligence regulatory decision-making. The study presents the overall landscape of artificial intelligence governance arrangements in the United States and measures the proximity of Sino-US regulation policies. This could serve as a vital reference for Chinese policymakers, providing evidence-based recommendations to guide the innovation and refinement of AI regulatory policies. It is critical for aligning China's regulatory strategies with global best practices, so increasing its ability to reduce risks, encourage ethical AI development, and boost its leadership in the global AI governance environment.

摘要： 捕捉美国人工智能监管动态并探寻中国参与竞争的政策安排是中国参与全球人工智能治理和提升全球科技领导力的关键议题。本文从政策复杂性视角，采用动态主题模型等多种分析技术，深入阐释了美国36个州的397份政策文本，呈现了人工智能监管的州域偏好，并测算了各州与中国人工智能监管的临近度。研究发现：美国人工智能监管政策系统雏形初现，产生了七个政策主题和四大核心政策工具。监管主题分布最丰富的州为加利福尼亚州、伊利诺伊州和纽约州；与中国监管政策最临近的是夏威夷州、俄克拉荷马州、纽约州；加利福尼亚州、马萨诸塞州等科技强州也在临近排名中居前列。最终，面向美国人工智能监管系统复杂性，提出了中国应采取“由近及远”、“化整为零”、“兼容共生”、“群策群力”的政策创新战略。