AI agent security
Detecting and localizing prompt-injection attacks in agent trajectories, and understanding when an agent resists or drifts from its intended task.
Northern Arizona University · Flagstaff, Arizona
Mithun Paul
Saint-Germain
Assistant Professor · School of Informatics, Computing, and Cyber Systems
I secure AI agents against adversarial attacks—and study how language and intelligence can become more trustworthy.
01 / About
Securing agents.
Building intelligence
we can trust.
I am an Assistant Professor at Northern Arizona University. AI agent security is a central focus of my research: how prompt injection and untrusted tool outputs change an agent's behavior, where attacks enter an agent trajectory, and how to detect and localize failures. My broader work spans natural language processing, fact verification, quantum natural language processing, and access to AI resources.
I lead KALI Lab—Knowledge, Adversaries, Language & Intelligence—where we investigate secure, trustworthy tool-using agents alongside language understanding and emerging computational approaches.
02 / Research
AI agent security
and beyond.
Secure autonomous agents lead a research program spanning language understanding and emerging computational methods.
Language & reasoning
Fact verification, knowledge distillation, causal reasoning, and more transferable language systems.
Quantum language
Compositional and multilingual foundations for quantum natural language processing.
03 / Selected publications
Selected work.
Follow each title to its individual paper record.
AgentDriftA step-labeled benchmark of injection-hijacked LLM agent trajectories↗
DriftNetDetecting and localizing prompt injection in LLM agents↗
Scaling Hindi Quantum Natural Language ProcessingAutomatic pregroup supertagging for scalable QNLP↗
AI-VERDEA gateway for egalitarian access to large language model resources↗
Phishing Email Detection Using Inputs From Artificial IntelligenceMachine learning for practical cyber defense↗
Data and Model DistillationA solution for domain-transferable fact verification↗
Complete publication record on Google Scholar ↗
04 / Teaching
Courses for a
thinking future.
Public course overviews for students exploring AI, deep learning, and responsible intelligent systems.
CS 102 · Fall 2026
Introduction to AI
AI literacy for a world shaped by intelligent systems.
Course overview
This interdisciplinary course builds a foundation in artificial intelligence, machine learning, data analytics, and responsible use. Students examine how AI systems work, where bias and privacy risks arise, and how AI is changing healthcare, transportation, creativity, and work.
Learning arc
- AI concepts: machine learning, neural networks, computer vision, and NLP
- Applications: search, recommendations, chatbots, assistants, and robotics
- Society: bias, fairness, privacy, accountability, and trust
- Practice: an ethical AI integration project in a domain of the student’s choice
Assessment
10%Attendance20%Quizzes30%Assignments40%Final project
CS 599 · Spring 2026
Deep Learning
Neural networks and natural language processing from first principles to practice.
Course overview
This course bridges the theory and practice of deep learning with applications to natural language processing. Students move from perceptrons and logistic regression to recurrent networks, transformers, language models, and practical systems built with PyTorch and Hugging Face.
Learning arc
- Foundations: perceptrons, logistic regression, and feed-forward networks
- Representations: distributional semantics and word embeddings
- Sequence models: recurrent networks, contextual embeddings, and transformers
- Applications: text classification, tagging, machine translation, and LLM inference
Assessment
20%Midterm 120%Midterm 225%Final exam10%Homeworks5%Paper readings20%Final project
Public overview only: private course logistics and student-facing materials are intentionally omitted.
05 / Curriculum vitae
A career in
trustworthy AI.
Selected academic, research, teaching, and industry experience.
Research profile
Ph.D. computer scientist working at the intersection of AI agent security, natural language processing, generative AI, quantum natural language processing, privacy-preserving machine learning, and fact verification.
Education
- Ph.D., Computer ScienceUniversity of Arizona · 2022
- M.S., Computer ScienceUniversity of Arizona · 2014
- M.S. Physics (Honors) & B.S. Engineering (Honors)BITS Pilani · 2005
Academic appointments
- Assistant Professor, Tenure TrackNorthern Arizona University · 2025–present
- Research Data ScientistUniversity of Arizona · 2023–present
- Postdoctoral ScholarInformation Sciences Institute, USC · 2022–2023
Industry experience
- Embedded Systems Software EngineerIntel · 2014–2015
- Software Engineer InternGoldman Sachs · 2013
- Technology ArchitectInfosys · 2005–2012
Invited talks
- Quantum NLP and artificial general intelligenceOxford · 2024
- Privacy-preserving large language modelsUniversity of Arizona · 2024
- Quantum NLP as a quantum advantageUniversity of Southern California · 2023
Service & recognition
- Conference reviewer and program committee memberACL, EMNLP, NAACL, COLING, LREC, FEVER, QNLP.ai
- Finalist, Three-Minute ThesisUniversity of Arizona · 2018
- Best research paper awardInternational Conference on Services in Emerging Markets · 2011
This is a selected public CV. Private contact details and internal course or workplace logistics are intentionally omitted.
06 / Contact
Let’s collaborate.
Interested in AI agent security, language research, or collaborating with KALI Lab? I would be glad to hear from you.