Making AI more trustworthy

Leon Swazinna has officially joined the Chair of Artificial Intelligence and Society as a PhD student under the supervision of Prof. Bilal Zafar. His work strengthens the team’s mission to better understand how artificial intelligence systems behave in real-world contexts - and how they can be made more reliable for society.

Leon is no stranger to the group. Already during his Master’s studies, he worked closely with the chair and completed his thesis under Prof. Zafar’s supervision. In his Master’s thesis, “Characterizing LLM Generations in Diverse Real-World Conditions,” Leon examined how large language models perform under extreme circumstances, such as unusually long inputs or outputs. His findings shed light on how such conditions influence the quality and robustness of AI-generated content - an issue of growing relevance as these systems are increasingly used beyond controlled laboratory settings.

Since mid-2025, Leon has also been contributing to the team as a part-time researcher. Following the completion of his Master’s degree, he joined the chair as a full-time PhD student at the beginning of this year, continuing his research at the intersection of large language models and society.

In his doctoral work, Leon focuses on two closely connected questions that are central to the chair’s broader research agenda on trustworthy AI. First, he investigates how we can determine whether a language model truly understands a task or question - rather than merely producing convincing but incorrect answers. By identifying such failures of understanding, his research aims to make AI systems more transparent and dependable.

Second, Leon explores the security risks that can arise when large language models are used in real-world applications. In collaboration with security researchers from another university, he works on identifying potential vulnerabilities and demonstrating how they may affect users. The goal is to contribute to long-term solutions that improve trust in AI systems and support their responsible use.

With his background, continuity within the team, and clear research focus, Leon Swazinna adds valuable expertise to the Chair of Artificial Intelligence and Society. His work contributes not only to academic progress, but also to informed decision-making, responsible innovation, and a broader public understanding of how AI systems shape - and should serve - society.