Robotics systems play an increasingly important role in armed conflicts and there are already weapons in service that replace a human being at the point of engagement. The United States (US) is the first country to have adopted a policy on autonomous weapon systems (AWS) in the Directive 3000.09. The US policy on AWS is however poorly understood in the academic and policy circles. This thesis addresses the question of how the US Department of Defense (DoD) problematises the concept of AWS. By applying a Bacchi’s poststructuralist approach to policy analysis, the thesis asks how the US DoD constructs the ‘problem’ of AWS, what assumptions underlie this representation of the ‘problem’, how has it come about, what effects it produces, what is left out of problem representation, and how could it be questioned. The US DoD problematisation of AWS does not only clarifies the Department’s approach, but also it explores the role of human involvement over the use of AWS. The US policy states that AWS shall be used by ‘appropriate levels of human judgment’. This term is, however, open to different interpretations, and some argue that it prohibits a lethal use of AWS, while others disagree. The thesis focuses not only on content of the US concept of human judgment, but primarily on how this concept relates to the wider US military understanding of ‘control.’ In that, it unpacks the concept of human judgment and distinguishes it from the concept of human control. I argue that both concepts are important in the debate on AWS as they represent alternative policy approaches to the use of such weapons. By making these 3 concepts more explicit, my thesis contributes to the specific and emerging academic debate about the role of human involvement over the use of AWS.