Anmol Goel

I’m an ELLIS PhD researcher at the UKP Lab at TU Darmstadt, supervised by Iryna Gurevych and co-supervised by Amartya Sanyal at the University of Copenhagen. My research focuses on the privacy and safety of large language models, particularly contextual privacy, agentic systems, and evaluation.

Recently, I have been focusing on building useful evaluation environments for LLM post-training and long-horizon capabilities. I’m especially interested in settings involving tool use, computer use, and multi-step interaction, where models must reason, act, and learn from feedback over extended tasks. More broadly, I study when language models and agents should disclose information, how privacy can fail across contexts, and how safety, alignment, and unlearning can improve model behavior after pre-training.

Previously, I spent two years at IIIT Hyderabad, working with Ponnurangam Kumaraguru. Outside of research, I enjoy cricket and travelling.