Imran Hossen

I am a tenure-track Assistant Professor in the Department of Computer Science and Engineering at the University of Tennessee at Chattanooga, where I joined in August 2025. Before that, I was a Postdoctoral Researcher in AI security at the University of Louisiana at Lafayette (2022–2025), where I also received my Ph.D. in Computer Science in 2022, advised by Dr. Xiali Hei.

My research is in offensive security for AI systems: I break them — models, agents, and the CAPTCHA gates in front of them — to understand how they fail and to build ones that don’t. Concretely:

• Security of large language models and AI agents — adversarial evaluation and red teaming of LLM-based systems, with a focus on code-generation and agentic workflows: jailbreaking, adversarial fine-tuning, and model-level compromise (backdoors, data poisoning) and their implications for software supply chain security.
• CAPTCHA attacks and design — machine-learning-based solvers against deployed human-verification systems (reCAPTCHA, hCaptcha, audio CAPTCHAs), and the design of next-generation schemes that stay usable while resisting adversarial ML — including verification anchored in the physical world.
• Physical and side-channel attacks on ML systems — attacks that exploit the physical world around a model, from acoustic side channels (e.g., precision keystroke tracking) to signal-injection attacks on biosensing pipelines, and defenses grounded in how these systems fail in practice.

Our CAPTCHA research has been covered by The Register and The Record.

Contact

echo bWRpbXJhbi1ob3NzZW5AdXRjLmVkdQ== | base64 --decode

Or: mdimran-hossen [at] utc [dot] edu