Advisor: David Krueger

Ekdeep Singh Lubana earned his PhD in EECS from University of Michigan in 2024. His PhD work focused on identifying and better understanding critical points that can cause risk regulation frameworks for AI to fail, e.g., sudden emergence of model capabilities, hindrances to risk assessments due to input underspecification, and inability of current posthoc interventions to patch model vulnerabilities. His recent interests are focused towards creating novel protocols for better evaluating open-ended models, demonstrating use of mechanistic interpretability as a risk monitoring toolkit, and methods for predicting agentic capabilities. Ekdeep will carry out his work at Center for Brain Science, Harvard University and will be co-advised by Hidenori Tanaka and David Krueger.