Ronald  Xu

Ronald Xu

Scholar Title

MIT Tang Family FinTech Undergraduate Research and Innovation Scholar

Research Title

Guardrails for LLMs Supporting Security





Research Areas
  • Cyber Security
  • Large Language Models
  • Reinforcement Learning

Una–May O'Reilly


The overall research goal is to experiment with different ways of training agents by prioritizing efficient training or high performance, to name a few. We will train several types of agents, such as goal-based agents and generative models, and apply them to the field of network security/cyber defense. For the generative AI agent, it is important that there are guardrails for the actions of the agent. We will use public cyber environment simulations to train and evaluate the agents, allowing them to learn and adapt to different scenarios while minimizing the potential risks and consequences associated with testing in real-world settings. Public cyber environment simulations will hopefully also ensure standardized and reproducible results.


I am participating in SuperUROP because I love doing research. Working on new, unsolved problems excites me and motivates me to work hard. My interest in research began in high school with science fair, and continued with UROP at MIT. I hope to gain a solid understanding of how to apply machine learning to the cybersecurity space and knowledge of how to train and evaluate generative AI that I can apply to other domains.

Back to Scholars