AI infrastructure security
The security and reliability of the stacks that train and serve models: accelerators, schedulers, and the shared data-center resources they contend for.
Ph.D. Student · Computer Science · University of Delaware
Zhicheng Sun
I study the security of AI infrastructure — the cloud platforms, data centers, and systems software modern AI depends on.
Understand how systems fail. Then make them harder to break.
My work sits where AI infrastructure, cloud, and systems security meet. I build systems, study how they fail, and look for practical ways to make them more secure.
Cloud & systems security
Isolation, resource management, and trust boundaries in multi-tenant cloud platforms and the systems software underneath them.
Vulnerability research
Large-scale measurement and coordinated disclosure of implementation flaws in web frameworks and language runtimes — 50+ CVEs assigned to date.
Current direction
Where AI workloads meet data-center power.
Large-scale training and inference produce sharp, correlated swings in power draw. I am studying how these workloads interact with data-center power management — and how that interaction can introduce security and reliability risks that neither layer sees on its own.
Cluster power
Background in web security, vulnerability research, healthcare cybersecurity, and software engineering, with industry experience in application security, cloud security, and compliance.
Selected work.
-
IEEE S&P 2026 Accepted
Poisoned by the Host: Large-Scale Measurement of Host Name Poisoning in Web Applications
My contribution: discovering vulnerability variants across language ecosystems and driving coordinated disclosure — 50+ CVE assignments, including high-severity issues in widely used web frameworks.
IEEE Symposium on Security and Privacy (S&P), 2026. Johns Hopkins University.
-
SIGCSE TS 2026 Poster
A Hands-On Platform for Medical Device Security Education
An open-source, secure-by-design medical device reference platform that turns FDA cybersecurity expectations and international standards into a reproducible, hands-on curriculum. Piloted in a graduate course at Johns Hopkins.
Proceedings of the 57th ACM Technical Symposium on Computer Science Education V.2, St. Louis, MO, February 2026.
Earlier work Clip2Sam: Enhanced End-to-End Text-to-Image Segmentation and Image Diffusion System. Zhicheng Sun, Yuan Jiang, Zihao Qin, Yancong Deng. 9th International Conference on Image, Vision and Computing (ICIVC), IEEE, 2024. DOI ›
Research and industry.
-
Present
Graduate Research Assistant
University of Delaware · X-Lab · Newark, DE
Research on AI infrastructure and systems security with Prof. Xing Gao, currently focused on how AI workloads interact with data-center power management and the security and reliability risks that follow.
-
2026
Graduate Researcher
Johns Hopkins University · Health & Medical Security Lab · Baltimore, MD
Vulnerability research and secure-by-design medical device systems. Co-authored an open-source reference platform for medical device security education, presented at SIGCSE TS 2026.
-
Aug 2025
Full-Stack Engineer, Intern
OfSpectrum · Remote
Owned vulnerability management and identity hardening for an AI audio-watermarking SaaS platform — remediated 10+ high-risk findings and cut unauthorized API calls by 85% in 90 days — and mapped platform controls to GDPR, SOC 2, and ISO 27001, reducing vendor-audit turnaround by 40%.
-
Jul 2024
Software Engineer, Intern
Inspur Group · Smart Healthcare Division · Jinan, China
Built an LLM-assisted EMR/EHR migration tool used across 100+ hospital migrations (85% mapping accuracy, ~70% less manual effort) and an OCR/NLP pipeline that produced a 100K+ labeled medical-entity dataset.
Foundations.
-
Ph.D., Computer Science
University of Delaware · X-Lab · Advisor: Prof. Xing Gao
-
M.S., Security Informatics
Johns Hopkins University · Information Security Institute
-
B.Eng., Software Engineering
Qingdao University
Let’s talk.
Happy to discuss AI infrastructure security, potential collaborations, or anything systems-related.