The use of Generative Artificial Intelligence (GenAI) in education is often framed as ‘cheating’ or ‘plagiarism’ while also raising fears of cognitive offloading, that is, the delegation of thinking and conceptualisation tasks to AI. Additional concerns pertain to errors in AI-generated content due to GenAI's limitations, particularly when used in smaller languages. To better understand the situation from a grassroots perspective, this chapter reports on a two-wave qualitative study that explores the experiences of MA students’ use of GenAI tools in one of their assignments. While uncovering a more interactive relationship between students and GenAI than much of the existing literature would suggest, implying that students retain significant agency in learning, this chapter also hints at deeper epistemic dependencies whereby not only the content but also the horizon of enquiry is tethered to the ‘black boxes’ of AI models, with a potential for tipping the balance of agency.