I’m interested in the robustness of machine learning models in real world settings. Currently, I am working on attacks and defenses on Computer Use Agents - systems that execute instructions over multiple steps and interact with your computer via screenshots to solve tasks under a changing environment. Ensuring their security and reliability is crucial, as such agents could cause substantial damage when dealing with sensitive data in high-impact workflows and not being properly safeguarded. Before starting my PhD, I completed my M.Sc. in Physics at RWTH Aachen University, where I used machine learning methods for quantum error correction.