Skip to content
arXiv· arXiv:2606.28450· Yiwei Xu, Yong Zhuang, Xuanming Liu, Tian Zhang, Bowen Xiao, Xiaoyang Xu, Delong Jiang, Juan Wang, Hongxin Hu·Indexed by this site · Originally published

The Security Duality of LLM Agents: A Comprehensive Survey of Self-Security and Empowering Cybersecurity

LLM agents security duality: a comprehensive survey of self-security and empowered cybersecurity

AI summary

An arXiv survey divides the security duality of LLM agents into "self-security" and "empowering cybersecurity": the former organizes internal and external attack surfaces, mitigation measures and evaluation frameworks by threat source, while the latter examines the application of agents across the full lifecycle of cyber offense and defense. The authors state that they propose the first agent empowerment framework aligned with the complete cyber offense and defense lifecycle, and are the first to point out that positive feedback synergy exists between the two parts. The article also lists current limitations and future research directions.

Read originalArtificial Intelligence Review