Research

Research Interests

My research centers on Secure AI Infra — securing modern AI systems across the entire stack, from silicon and accelerators to foundation models and autonomous agents — with three primary thrusts:

1. Model & Data Security — Protecting ML models from extraction, probing, and misuse; defending against adversarial, poisoning, and backdoor attacks
2. Hardware Security & Confidential Computing — Side-channel analysis (EM, power, cache) of AI systems and TEE-based protection for secure model inference
3. Agentic AI Security — Securing LLM and Mixture-of-Experts deployments and agentic AI platforms (e.g., vibe coding, openclaw), including side-channel threats to inference infrastructure

These efforts advance one mission: building trustworthy AI infrastructure — from silicon and accelerators to foundation models and autonomous agents. Now I am actively working on security of agentic AI, including vibe coding, openclaw, etc.