RESUMO
For two decades scholars have used insights from constructivist approaches and principal-agent (P-A) theory to understand the relationship between states and international organisations (IOs). Together, these works identify the conditions under which IOs can operate independently of states, although they have yet to explain when and why IO bureaucrats are likely to do so. Nor do they articulate a clear and consistent definition of autonomy. In this article, we seek to fill these gaps. We advance a narrow understanding of autonomy that distinguishes unintended behaviour from the intended independence of IO bureaucrats, before developing a three-stage, integrative explanation for the conditions under which IO bureaucrats act autonomously. First, we borrow from constructivist approaches a focus on staffing rules and the identity of IO bureaucrats to explain the sources of these agents' preferences. Second, we add insights from work on exogenous pressures for change-crises and critical junctures-to explain when and why IO bureaucrats will advance their preferences. Third, we incorporate P-A theory's attention to an IO's institutional design, along with insights from literature on domestic institutions, to explain when bureaucrats can implement their preferences. Case studies of the World Health Organization (WHO) and the World Trade Organization (WTO) illustrate our argument.