Assistant Professor Song Liao’s two-year project “CyberAI Innovation: Securing and Leveraging Agentic AI for Cyber Defense” focuses on training students at the intersection of agentic artificial intelligence (AI) and cyber security.
Song Liao, an assistant professor of computer science in the Edward E. Whitacre Jr. College of Engineering was awarded $249,848 from the National Science Foundation to prepare students to leverage AI agents for cyber defense while understanding the security challenges associated with these systems.
The rapid advance of AI technology is transforming many aspects of everyday life. Large language models (LLM) like Chat GPT and Gemini are now used for a wide range of activities from basic searches to supporting complex tasks such writing software, analyzing large datasets, and assisting with scientific research.
Liao’s work will focus on the more complex AI agent, which is powered by large language models, but that can reason and plan and can automatically cause external tools to perform tasks. While these agents can be very helpful, they can pose new risks because they may access data, use software tools, and take actions in ways that are difficult to predict or control.
“It is critical that cyber security keeps pace with AI development, and the best way to do that is to train computer scientists in both disciplines,” explained Liao.
Liao and his collaborating principal investigator Long Cheng, an associate professor at Clemson University, will first develop a curriculum that trains computer science students to approach agentic AI and cyber defense from both sides of the framework. As AI advances at a rapid pace, cyber security professionals must have a robust background in agentic AI, and those professionals developing agentic AI need to understand the security impacts of this emerging technology.
“AI agents are creating tremendous opportunities, but they also introduce new cybersecurity challenges,” said Yong Chen, chair of the Department of Computer Science. “This project is an important step toward preparing our students to both leverage these emerging technologies and ensure they are developed and used securely.”
The curriculum will include modules in AI agents, core cybersecurity principles, AI agents for security learning, and security in AI agents. Students will be prepared to identify a wider range of critical issues that can occur in cybersecurity.
Once students have completed these modules, they will engage in hands-on labs that leverage AI agents as interactive learning tools and investigate security challenges in real-world AI agent systems. After completing lab-based testing scenarios, students will complete projects to design and evaluate AI-agent based defenses against different attacks.
Liao’s work will also evaluate how students interact with large language model tools, an area of limited study to date.