Yuepeng Hu's Homepage

About

I am a fourth-year Ph.D. candidate in Electrical & Computer Engineering at Duke University, advised by Prof. Neil Gong. My research studies the security and safety of generative AI — spanning LLM agents and prompt injection, watermarking and detection of AI-generated content, safety of (multimodal) large language models, and trustworthy machine learning.

I am currently a Research Scientist Intern at Meta Superintelligence Labs, where I study reward hacking in RL-trained coding agents. Previously, I was a Research Scientist Intern at TikTok Privacy Innovation Labs. I received my B.S. in Data Science from the University of Science and Technology of China (USTC).

I am actively seeking full-time opportunities — please feel free to reach out.

Industry Experience

  • Research Scientist Intern Meta Superintelligence Labs · Bellevue, WA
    May 2026 – Present
  • Research Scientist Intern TikTok Privacy Innovation Labs · San Jose, CA
    May 2025 – Aug 2025

Education

  • Ph.D. in Electrical & Computer Engineering Duke University
    Aug 2022 – Present
  • B.S. in Data Science University of Science and Technology of China (USTC)
    Aug 2018 – Jun 2022

News

  • Started as a Research Scientist Intern at Meta Superintelligence Labs (Bellevue, WA).
  • Two papers accepted to ACL 2026: “Fingerprinting LLMs via Prompt Injection” (Oral) and “Leave My Images Alone”.
  • “Watermark-Based Detection and Attribution of AI-Generated Content” accepted to ICLR 2026.
  • “Jailbreaking Safeguarded Text-to-Image Models via LLMs” accepted to EACL 2026 (Findings).
  • “WebInject: Prompt Injection Attack to Web Agents” accepted to EMNLP 2025 (Main).
  • Started as a Research Scientist Intern at TikTok Privacy Innovation Labs (San Jose, CA).
  • “Periodic Recovery from Poisoning Attacks in Machine Learning” accepted to IEEE TDSC (IEEE Transactions on Dependable and Secure Computing).
  • “A Transfer Attack to Image Watermarks” accepted to ICLR 2025.

Selected Papers

View all →

* denotes equal contribution · my name in bold.

Fingerprinting LLMs via Prompt Injection ORAL

Yuepeng Hu, Zhengyuan Jiang, Mengyuan Li, Osama Ahmed, Zhicong Huang, Cheng Hong, Neil Gong

ACL (Annual Meeting of the Association for Computational Linguistics), 2026

arXiv Code

Proposes a prompt injection technique to uniquely fingerprint and track large language models.

MalTool: Malicious Tool Attacks on LLM Agents

Yuepeng Hu, Yuqi Jia, Mengyuan Li, Dawn Song, Neil Gong

Preprint, 2026

arXiv

Investigates vulnerabilities in LLM agents when interacting with external tools, demonstrating malicious tool-based attacks.

A Transfer Attack to Image Watermarks

Yuepeng Hu, Zhengyuan Jiang, Moyang Guo, Neil Gong

ICLR (International Conference on Learning Representations), 2025

arXiv Code

Demonstrates cross-model transferability of attacks against image watermarks, highlighting fundamental robustness issues.