Project Description

Modern AI systems increasingly work in teams, where multiple agents collaborate to solve complex tasks. In this project, you will study what happens when one or more AI teammates behave maliciously, either due to bugs, attacks, or intentional manipulation, and how to design cooperative AI systems that remain reliable and safe in these settings. You will work with large language model (LLM)-based agents (e.g., OpenClaw agents) in multi-agent environments and explore how insider attacks can disrupt coordination, decision-making, and trust. Together with Dr. Zheng and his PhD students, you will implement both attack strategies and defense mechanisms, run experiments on multi-agent tasks, and analyze system behavior under adversarial conditions. This project is supported by a National Science Foundation (NSF) grant through an REU supplement.

What You Will Do

• Implement multi-agent environments with cooperative AI agents
• Design and test insider attack strategies (e.g., deceptive or sabotaging agents)
• Develop and evaluate defenses that improve system robustness
• Run experiments, analyze results, and visualize agent behavior
• Contribute to writing a research paper based on your findings

What You Will Gain

• Hands-on experience with multi-agent AI and LLM-based agents
• Exposure to AI security, robustness, and trustworthy AI research
• Strong Python and experimental research skills

• Close mentorship from a faculty member and PhD students
• Opportunity to co-author a peer-reviewed research paper

Project Outcome

By the end of the project, you will have helped build and evaluate attack and defense techniques for cooperative AI systems, contributed to reusable experimental tasks and datasets, and participated in preparing a research paper summarizing the results.

Project Details

Time, eligibility, and other details