Pawelczyk, Martin

My research focuses on the safety and reliability of artificial intelligence within the broad fields of AI Safety and Data-Centric AI. My goal is to develop machine learning methods that bridge the gap between technological capabilities and ethical responsibility. Specifically, I work on methods and evaluation frameworks to make AI models - particularly Large Language Models (LLMs) - safer, more interpretable, and more privacy-preserving. In doing so, I address fundamental questions: How can we reconcile privacy and utility? What influence does data curation have on model safety? And how can complex AI decisions be made comprehensible for humans?

• Privacy Frameworks and Machine Unlearning (balancing privacy and utility)
• Impact of data curation on the safety of AI models
• Interpretability (Explainable AI) to assist human decision-making
• Scalable evaluation methods for generative models and Large Language Models
   

• 2013-2016 BSc in Economics at the University of Cologne and the Université du Québec à Montréal (UQAM), Canada
• 2016-2017 MSc in Econometrics (Applied Statistics) at the University of Edinburgh, UK
• 2017-2018 MSc in Statistics at the London School of Economics and Political Science (LSE), UK
• 2018-2023 PhD in Computer Science at the University of Tübingen
• 2021-2023 Research Fellow at Harvard University, USA
• 2022-2023 Research Fellow at the Max Planck Institute for Security and Privacy (Bochum)
• 2023-2025 Postdoctoral Fellow at Harvard University, USA
• since Feb 2026 tenure track professorship in Responsible Computing, University of Vienna