In this article I problematize the use of algorithmic decision-making (ADM) applications to automate legal decision-making processes from the perspective of the European Union (EU) policy on trustworthy artificial intelligence (AI). Lately, the use of ADM systems across various fields, ranging from public to private, from criminal justice to credit scoring, has given rise to concerns about the negative consequences that data-driven technologies have in reinforcing and reinterpreting existing societal biases. This development has led to growing demand for ethical AI, often perceived to require human control over automation. By engaging in discussions of human-computer interaction and in post-structural policy analysis, I examine EU policy proposals to address the problematizations of AI through human oversight. I argue that the relevant policy documents do not reflect the results of earlier research which have undeniably demonstrated the shortcomings of human control over automation, which in turn leads to the reproduction of the harmful dichotomy of human versus machine in EU policy. Despite its shortcomings, the emphasis on human oversight reflects broader fears surrounding loss of control, framed as ethical concerns around digital technologies. Critical examination of these fears reveals an inherent connection between human agency and the legitimacy of legal decision-making that socio-legal scholarship needs to address.