PhD student, University of North Carolina at Chapel Hill
1 paper at NeurIPS 2025
We introduce the "Deadlock Attack," which uses an adversarial embedding to force LLMs into perpetual thinking loops, leading to resource exhaustion.
