Mahavir Dabas
Mahavir Dabas is a Ph.D. student in the Bradley Department of Electrical and Computer Engineering. His advisor is Ruoxi Jia.
In his research, Dabas is particularly interested in the safety, security, and alignment of artificial intelligence models across different settings and modalities. Currently, he is looking into adversarial robustness against jailbreak attacks and developing methods to improve model alignment while reducing over-refusal in safety-tuned systems.