Agentic AI has the potential to transform scientific research into AI-mediated research, in which intelligent agents perform many cognitive tasks in the research lifecycle. While this development may accelerate scientific discovery, it also creates a verification gap between machine-generated knowledge and the human capacity to validate it. We propose that Open Science and the knowledge infrastructure managed by academic libraries are part of the solution to address this verification gap. Open Science may become even more important with the focus shifting from openness of research outputs towards transparency and reproducibility of the research process. In this context, the knowledge resources managed by academic libraries – whose users may increasingly be AI agents – could become a bastion of publicly governed knowledge infrastructures essential for academic sovereignty. In this paper, we present the possible effects of AI-mediated research on scientific knowledge production, on Open Science and on the role and services of academic libraries.