cv
Updated on August 18th 2026.
Basics
| Name | Zedian Shao |
| Label | Ph.D. Student in Computer Science |
| zedian.shao@outlook.com | |
| Phone | +1 919-672-7127 |
| Url | https://asympsec.com |
| Summary | My research focuses on AI security and safety, especially attacks and defenses for generative AI. I am currently interested in mechanistic interpretability as a tool for understanding, evaluating, and hardening AI systems against security and safety failures. |
Education
-
2025 - Present Atlanta, GA
Ph.D.
Georgia Institute of Technology, School of Cybersecurity and Privacy
Computer Science
- Advisor: Prof. Teodora Baluta
-
2023 - 2025 Durham, NC
M.Sc.
Duke University, Pratt School of Engineering
Electrical and Computer Engineering
- Advisor: Prof. Neil Zhenqiang Gong
- GPA: 4.0/4.0
-
2019 - 2023 Suzhou, China
Interests
| AI Security and Safety | |
| Generative AI Attacks and Defenses | |
| Mechanistic Interpretability | |
| Security and Safety Evaluation | |
| AI System Hardening |
Publications
-
2026 A Critical Evaluation of Defenses against Prompt Injection Attacks
ACM Symposium on Access Control Models and Technologies (ACM SACMAT)
Yuqi Jia, Zedian Shao, Yupei Liu, Jinyuan Jia, Dawn Song, and Neil Zhenqiang Gong.
-
2026 PromptLocate: Localizing Prompt Injection Attacks
IEEE Symposium on Security and Privacy (S&P)
Yuqi Jia, Yupei Liu, Zedian Shao, Jinyuan Jia, and Neil Zhenqiang Gong.
-
2026 Leave My Images Alone: Preventing Multi-Modal Large Language Models from Analyzing Images via Visual Prompt Injection
Annual Meeting of the Association for Computational Linguistics (ACL)
Zedian Shao* and Hongbin Liu*, with Yuepeng Hu and Neil Zhenqiang Gong. *Equal contribution. Code: https://github.com/Sadcardation/ImageProtector
-
2026 Cordyceps: Covert Control Attacks on LLMs via Data Poisoning
USENIX Security Symposium (USENIX Security)
Zedian Shao, Charles Fleming, and Teodora Baluta. Code: https://github.com/Sadcardation/cordyceps
-
2025 WebInject: Prompt Injection Attack to Web Agents
Conference on Empirical Methods in Natural Language Processing (EMNLP)
Xilong Wang, John Bloch, Zedian Shao, Yuepeng Hu, Shuyan Zhou, and Neil Zhenqiang Gong.
-
2025 Enhancing Prompt Injection Attacks to LLMs via Poisoning Alignment
ACM Workshop on Artificial Intelligence and Security (AISec)
Zedian Shao* and Hongbin Liu*, with Jaden Mu and Neil Zhenqiang Gong. *Equal contribution. Code: https://github.com/Sadcardation/PoisonedAlign
Work
-
2025.08 - 2025.12 Atlanta, GA
Teaching Assistant | CS 8803: Machine Learning Security
Georgia Institute of Technology, College of Computing
- Designed assignments on adversarial examples, backdoor and poisoning attacks, watermarking, and formal-methods topics for a graduate machine learning security course.
- Graded written and programming assignments, developed feedback rubrics, and held office hours to help students reason through attacks, defenses, and implementation details.
-
2023.01 - 2023.04 Durham, NC
Teaching Assistant | DECISION 521Q: Decision Analytics and Modeling
Duke University, The Fuqua School of Business
- Evaluated analytics and modeling assignments, clarified optimization and decision-analysis concepts, and supported students through office hours.
-
2022.09 - 2023.06 Singapore
Software Development Engineer
Lexikat Pte. Ltd.
- Built the core architecture of an iOS privacy-protection application in SwiftUI, integrating PhotosUI and UIKit for media selection, preview, and editing workflows.
- Implemented efficient image and video processing pipelines with PhotoKit and Image I/O to support privacy-preserving transformations on user media.
- Developed OpenCV-based text-image transformation features designed to reduce OCR readability while preserving human-readable content.
Projects
-
AI Security Research Knowledge Base Telegram Bot
Python, FastAPI, LangGraph, Qdrant, PostgreSQL
- Built a personal Telegram bot-based AI-security research assistant with an agent harness to coordinate tool execution, workflow state management, retrieval pipelines, and multi-step research reasoning.
- Designed a multi-agent workflow for topic classification, query expansion, source retrieval, grounded synthesis, and conversation summarization.
- Implemented a RAG knowledge-base pipeline for arXiv PDFs, conference papers, GitHub READMEs, and personal notes using semantic chunking, metadata filtering, hybrid search, and reranking.
-
Toy Rootkit: Stealth Kernel Module for Process Concealment
C, Linux Kernel Programming, System Call Hooking
- Built a Linux kernel module that intercepts system calls to support process hiding, module hiding, and file access redirection.
- Modified the sys_call_table to hook getdents64, openat, and read, concealing target processes and kernel modules from user-space utilities such as ls, ps, and cat /proc/modules.
-
Daemonized Multithreaded HTTP Caching Proxy
C++, Boost, Web Programming, Multithreading, Daemon
- Implemented a daemonized C++/Boost HTTP caching proxy that parses client requests, forwards upstream traffic, and stores reusable responses.
- Designed an LRU-based cache with HTTP revalidation logic to reuse cached resources while preserving response freshness.
- Added multithreaded request handling to serve concurrent client connections and isolate long-running network operations.
Awards
- 2026
USENIX Security Student Grant and GREPSEC VII Travel Award
USENIX Security and GREPSEC VII
- 2022
Dean's List with Distinction (top 10%)
Duke Kunshan University
- 2019
Merit-based Scholarship
Duke Kunshan University
- 2018
Second Prize, Chinese Chemistry Olympiad (CChO)
Chinese Chemistry Olympiad
Volunteer
-
2026.06 - 2026.06 -
2025.05 - 2025.05 -
2023.07 - 2023.07
Skills
| Programming Languages | |
| Python | |
| C/C++ | |
| Verilog |
| Tools and Frameworks | |
| Git | |
| LaTeX | |
| PyTorch | |
| Docker | |
| Django | |
| PostgreSQL |