Lethal Autonomous Weapons Systems (LAWS) have the potential to dramatically alter modern warfare and reshape global power differentials. Despite the strong rationale for negotiating global rules, consensus on whether and how to regulate LAWS has yet to be reached. This article explains multilateral bargaining over global LAWS regulation. Theoretically, it draws on rational choice institutionalism to develop an argument about the temporal dynamics of the negotiations over LAWS regulation. I emphasize the role of great power preferences and coalition formation in impeding preference convergence. Empirically, it focuses on the principal forum discussing LAWS regulation—the Group on Governmental Experts (GGE) under the Convention on Certain Conventional Weapons (CCW)—and uses a novel dataset, complemented with text analysis methods to test the theoretical framework. The findings are threefold. First, several great powers consistently oppose new regulations on LAWS, distinguishing them from most other negotiating parties advocating for stricter rules. Second, weaker states turn to coalition formation as a strategy to augment their power. Third, these two factors jointly hinder the concession-convergence dynamic typically expected under consensus decision rules. This article has implications for debates on the regulation of LAWS, arms control negotiations, and the global governance of AI.

Which states shape emerging international regulation of artificial intelligence (AI) and why? The European Union (EU) AI Act represents the first major international framework on AI, adopted in a context where governments worldwide increasingly pursue regulatory coordination. Its adoption followed a lengthy negotiation process marked by periods of intense contestation, the addition of thousands of amendments, and growing time pressure, providing a window into the politics of international AI regulation. This article uses the EU AI Act as a laboratory to study state influence over AI regulation and offers the first systematic analysis of these negotiations. Theoretically, we develop competing expectations on the role of AI competitiveness for state influence. Empirically, we construct an original dataset to conduct a preference attainment analysis, estimating states' bargaining influence by comparing their initial policy positions with the negotiated outcome on four key issues of contestation. We find that AI competitiveness is negatively related to bargaining influence, suggesting that states most dependent on regulatory harmonization may be less able to shape outcomes. These findings advance our understanding of the politics of the AI Act before and after its adoption, contribute to the growing literature on AI regulation, and have implications for research on EU decision-making.

The promises and pitfalls of artificial intelligence (AI) have led to an intense race to regulate the technology. However, no central international regulatory institution has yet emerged. Instead, the regulatory landscape is characterized by a variety of nascent and partially overlapping governance arrangements, resembling an emerging regime complex. What are the key characteristics of this regime complex, and what logics can account for its origin and shape? This article offers the first systematic analysis of the AI regime complex. Its contribution is twofold. First, theoretically, we develop three explanatory logics—power-oriented, functionalist, and pathdependent— to evaluate the evolution and shape of the AI regime complex along two key structural dimensions: hierarchy and differentiation. Second, empirically, we track and explain the development of the AI regime complex 2014–2024 based on a novel and comprehensive dataset on 119 international institutions involved in AI governance. Our findings show that the AI regime complex is marked by a lack of hierarchy, dense geographic concentration of governance efforts in Europe, and generally low functional differentiation, reflecting the combined influence of state power, functional demands, and institutional path dependencies. These patterns point to risks of fragmentation in global AI governance, as overlapping institutions and regional concentration may impede coherent rule-making and constrain the regime complex’s capacity to adapt to emerging challenges.

Artificial Intelligence (AI) is widely recognized as a transnational governance challenge, but global AI governance remains fragmented and largely non-binding, with current efforts deemed inadequate. Yet, little is known about what makes for effective AI governance. I introduce a novel conceptualization of precision as a relevant dimension of AI policy: legal and technical precision. While legal precision relates to a policy's legal clarity and prescriptiveness, technical precision denotes the extent to which it specifies the technical properties and complexity of AI systems. I expect legal and technical precision to vary across the policies produced by different actors in global AI governance; to examine this, I focus on public, public-private, and private actors. I test my argument on a newly collected dataset on global AI governance documents, utilizing a dictionary-based measure of legal and technical precision in descriptive and explanatory analysis. The findings support this conceptualization of precision. Actor type correctly predicts differences in technical precision, yet the results for legal precision run counter to expectations. This paper contributes to the literature on precision as a dimension of institutional design, while introducing a new measurement approach that speaks directly to the effectiveness of AI policies, thereby advancing contemporary debates on global AI governance.