Knowledge · Adversaries · Language · Intelligence
Securing AI agents.
Advancing trustworthy intelligence.
AI agent security is at the center of our work. We study how tool-using agents resist prompt injection and behavioral hijacking, how intelligent systems understand language, and how advanced AI can become more accessible.
Explore our research
KALI—Knowledge, Adversaries, Language & Intelligence—is led by Dr. Mithun Paul Saint-Germain. Our central mission is to secure AI agents: identify attacks in their trajectories, understand when behavior drifts, and protect the actions they take. This work connects to language technology and emerging computational paradigms. We care about agents that work beyond the benchmark—under adversarial pressure, untrusted tool outputs, and real-world access constraints.
KALI Lab Research
The shield is our research promise: observe what an agent sees, understand what it trusts, and protect the actions it takes.
Three connected directions
Securing tool-using agents against indirect prompt injection: identifying where an attack enters, which actions it corrupts, whether the agent resisted, and how its behavior drifted.
Language understanding and reasoning for agents: fact verification, causal models, and data-efficient transfer across domains.
Scalable quantum natural language processing, compositional models, and multilingual grammar-aware foundations for future intelligent agents.
Agent security · language · intelligence
A collaborative agent research group
Our work brings together AI-agent security, natural language processing, cybersecurity, and emerging computational methods. We are especially interested in students and collaborators who want to make autonomous, tool-using agents safer under adversarial pressure.
Principal investigator
Send a short introduction, your agent research interests, and a CV or résumé.
Start an agent security conversation
School of Informatics, Computing, and Cyber Systems
Northern Arizona University · Flagstaff, Arizona