Tianhao Li
Graduate Student, Duke University; Founder & Head of Research, SIGMIR
Durham, North Carolina, United States
Actions
Tianhao Li is a graduate student at Duke University, where he work with Prof. Neil Gong, Prof. Fang-Fang Yin, and Prof. Zhenyu Yang. He is also a visiting researcher at the SaFoLab, University of Wisconsin-Madison, work with Prof. Chaowei Xiao. His research aims to evaluating and enhancing the safety and privacy of generative models and systems, particularly in real-world applications such as healthcare, science, and the metaverse. He received a B.Eng. in Information Security in 2024 from North China University of Technology and worked as a Security Researcher (AI Red Teaming) at NSFOCUS and TOPSEC during his undergraduate studies. He also serves as peer reviewers for prestigious journals and conferences, including TIST, TAI, TBE, EAAI, RESS, JBHI, AAAI, IJCAI, ICLR, and ACL, etc.. In addition, he contribute to the MLCommons AI Risk & Reliability (AIRR) Working Group, and the NVIDIA's widely recognized open-source project NVIDIA/garak. In early 2025, he founded Special Interest Group in Modern Interdisciplinary Research (SIGMIR) under section 501(c)(3), built its core team, and led its early operations and strategic development.
Area of Expertise
Red Teaming Agent-to-Agent (A2A) Multi-Agent Systems: Threat Modeling and Protocol Exploitation
This session introduces red teaming strategies for Agent-to-Agent (A2A) multi-agent systems, a new protocol for secure agent collaboration. We dissect the A2A lifecycle—capability discovery, task delegation, collaboration, and completion—and reveal how each phase presents unique security threats. Attendees will learn how to identify vulnerabilities, simulate attacks such as AgentCard poisoning, task lifecycle manipulation, and adversarial MIME payloads, and evaluate system resilience using a structured threat model. This session is ideal for researchers and practitioners building or defending agent-based AI systems.
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top