Guest Editor(s)
Prof. Shen Su
Email: sushen@gzhu.edu.cn
Affiliation: School of Cyberspace Security, Guangzhou University, Guangzhou, China
Homepage:
Research Interests: cyber attack and defense, blockchain security
Dr. Donglei Wu
Email: dowu@gzhu.edu.cn
Affiliation: School of Cyberspace Security, Guangzhou University, Guangzhou, China
Homepage:
Research Interests: distributed machine learning, AI security
Summary
Artificial Intelligence (AI) is rapidly evolving from passive models to autonomous agent systems that can interact with tools, environments, and users. Powered by large language models and tool-use capabilities, these AI agents are now used in applications such as intelligent assistants, automated workflows, and even cyber operations, where they can plan and execute multi-step tasks. However, this increased capability also brings new security risks beyond traditional AI systems. When agents exchange prompts, intermediate reasoning steps, or tool outputs, they may unintentionally expose sensitive information or become vulnerable to manipulation. Traditional security methods, which are designed for static models and fixed system boundaries, are often not sufficient in such dynamic and interactive settings. As a result, new threats emerge, including prompt injection, tool misuse, broken trust boundaries, data leakage, and adversarial attacks on agents. These risks can spread across multiple steps in an agent workflow, making the impact more severe. Therefore, ensuring the security and reliability of AI agent systems is becoming increasingly important as they gain more autonomy and play a larger role in real-world decision-making.
This Special Issue aims to investigate the emerging security challenges and defense mechanisms in AI agent systems. It focuses on the unique attack surfaces introduced by agent autonomy, including tool invocation, external data interaction, and execution workflows. Topics of interest include prompt injection attacks, tool and API misuse, adversarial manipulation, secure agent architectures, and trust management. The issue welcomes both theoretical and practical contributions that advance the understanding, design, and evaluation of secure and resilient AI agents in real-world applications.
Suggested themes include, but are not limited to:
· Threat modeling and attack surface analysis for AI agents
· Prompt injection and adversarial attacks in AI agents
· Secure tool use and API interaction in agent systems
· Trust, alignment, and controllability of autonomous agents
· Secure memory, session, and context management
· Agent autonomy and risk-aware decision-making
· Detection and mitigation of malicious or compromised agents
· Identity, authentication, and access control for AI agents
· Confidentiality, integrity, and availability in agent systems
· Privacy-preserving agent learning and interaction
· Multi-agent interaction security and coordination risks
· Human-Agent Interaction and Governance Mechanisms
· Secure planning and reasoning in long-horizon agent tasks
· Secure Agent Deployment and System Architecture
· Security evaluation, auditing, and benchmarking of AI agents
Keywords
AI agents, AI security, prompt injection, tool use security, trustworthy AI
Login
Register
Submit a Paper
Propose a Special lssue