Bio

Meriel von Stein is an Information Scientist at RAND Corporation, where she studies questions at the intersection of AI reliability, safety, and technology policy. Her work brings technical grounding to policy-relevant questions about how AI and autonomous systems behave, fail, and should be governed.

She holds a Ph.D. in Computer Science from the University of Virginia, where she was advised by Dr. Sebastian Elbaum and was a founding member of LESS Lab. Her doctoral research developed techniques for assessing and ensuring the robustness of AI/ML-enabled safety-critical systems, with applications to autonomous vehicles and robotics. Her work has been published in leading venues including Transactions on Software Engineering (TSE), the IEEE International Conference on Software Engineering (ICSE), IEEE Robotics and Automation Letters (RA-L), and the International Conference on Robotics and Automation (ICRA). She was recognized as a Copenhaver Graduate Fellow and received UVA’s university-wide Graduate Teaching Award.

Before her PhD, Meriel worked as a software engineer at NASA Kennedy Space Center and NASA Goddard Space Flight Center, contributing to launch control software testing and satellite ground control systems development.

Meriel is an active member of the research community. She has co-chaired workshops at top venues (FSE 2023), reviews regularly for leading conferences and journals, and has mentored undergraduate and high school students on research projects, including FIRST Robotics competitions.

Research

Meriel’s research sits at the boundary of software engineering, AI/ML, and policy. She is interested in how AI systems fail in the real world — particularly in safety-critical applications — and what that means for how we build, evaluate, and regulate them.

Her technical work focuses on weaknesses in learned perception components, the nondeterministic behavior of autonomous systems in physical environments, and methods for testing and analysis that are meaningful under uncertainty. She is increasingly interested in how these technical realities translate into policy questions: what can regulators and decision-makers actually know about AI system reliability, and what evidence standards are both rigorous and tractable?

Prior to RAND, this work was situated in robotics and autonomous vehicles, where the stakes of perception failures are high and the systems are difficult to evaluate exhaustively.