Topic: Trustworthiness and Security Risks of Agents in Complex Open Environments, and Their Mitigation
Speaker: Asst. Prof. Bo Wang (School of Artificial Intelligence, Jilin University)
Abstract
With the rapid improvement of large language models in capabilities such as reasoning, planning, tool use and long-term memory, artificial intelligence is gradually evolving from large models that mainly perform content generation and question-answering interaction toward agents that can perceive their environment, make autonomous decisions and execute complex tasks. The growing autonomy and task-execution capability of agents bring new opportunities for applying AI in complex real-world scenarios, while also extending the trustworthiness and security issues they face from the level of model outputs to multiple stages, including information acquisition, reasoning and decision-making, long-term memory, and action execution. Existing agents, however, still face problems such as the difficulty of guaranteeing the reliability of external information, the vulnerability of reasoning and planning processes to malicious information, privacy leakage and malicious manipulation risks in long-term memory, and security threats that may arise from tool use and autonomous behavior. To address these issues, this talk presents relevant research advances in the field of agent trustworthiness and security. Drawing on related research and representative cases, it analyzes the typical trustworthiness and security risks that agents face in complex open environments, introduces related attack methods, their underlying mechanisms and potential mitigation strategies, and offers new research perspectives for understanding agent trustworthiness and security issues and for developing secure and reliable agent technologies.
Speaker Bio
Bo Wang is an Assistant Professor and doctoral supervisor at the School of Artificial Intelligence, Jilin University. His research focuses mainly on trustworthy artificial intelligence and natural language processing, with interests including the security and privacy of large models and agents, retrieval-augmented generation, and misinformation detection. He has published more than ten papers in well-known CCF Category A/B international conferences and journals such as ACL, WWW, EMNLP and TOIS, with over one thousand citations on Google Scholar in total. He serves as a reviewer for internationally renowned journals and conferences including TPAMI, TKDE, ACL and EMNLP. He leads one project under the National Natural Science Foundation of China (NSFC) Young Student Basic Research Program.
Time: Wednesday, November 4, 2026, 15:30–17:10
Venue: Undergraduate Major Classroom, 4th Floor, Zhengxin Building, Qianwei Campus, Jilin University
