Jianwei Li

Hi! I am currently a PhD student at North Carolina State University, under the guidance of Prof. JUNG-EUN KIM. Previously, I worked as a machine learning scientist in Moffett AI, and was guided by the chief scientist and co-founder of Moffett.AI: Ian En-Hsu Yen. Before that, I got my master's degree from the CS department of San Jose State University in 2021 and was advised by Prof. Mark Stamp. I earned my bachelor's degree from Shandong University in 2016.

I will be joining the TikTok NLP Trust & Safety (TNS) team as a Research Scientist Intern in Summer 2026. Previously, in Summer 2025, I interned with TikTok Responsible Recommendation Systems (RRS) team, where I worked on safety-aware recommendation for multimodal LLM.

CV / Google Scholar / LinkedIn / Twitter

Shadow LLM Guardians

Address: Raleigh, North Carolina
Email: ljw040426 AT gmail DOT com | jli265 AT ncsu DOT edu
Research Interest

My researches primarily focuses on the field of AI Safety & Security and AI Efficiency, especially in the topic of Alignment and Attacks/Defenses.

News
* Scrollable
  • [Jan, 2026] Two papers are accepted by ICLR 2026
  • [Oct, 2025] Pass PhD written preliminary exam at NCSU
  • [May–Aug, 2025] Start a research internship program at TikTok
  • [May, 2025] One paper is accepted by ICML 2025
  • [Feb, 2025] One paper is accepted by CPAL 2025
  • [Jul, 2024] Start research under the guidance of Prof. Jung-Eun Kim
  • ----------------------------------------
  • [Dec, 2023] One paper selected as Oral of FL@FM-NeurIPS 2023
  • [Nov, 2023] Initiated the Shadow-LLM-Guardians Group: https://github.com/Shadow-LLM
  • [Oct, 2023] Two papers are accepted by EMNLP 2023
  • [Aug, 2023] Start PhD student life at NC State University; Serve as official affiliation of New York University
  • ----------------------------------------
  • [Apr, 2023] Publicity Chair of the International Workshop onResource-Efficient Learning for Knowledge Discovery @KDD 2023.
  • [Oct, 2022] Publicity Chair of the first workshop on DL-Hardware Co-Design for AI Acceleration @AAAI 2023.
  • [Sep, 2022] Participated in AI Hardware Summit 2022.
  • [Sep, 2022] Moffett S30 Accelerator wins MLPerf V2.1
Selected Publications
pub
Jianwei Li, Jung-Eun Kim
Purifying Generative LLMs from Backdoors without Prior Knowledge or Clean Reference
Main PaperDefense Backdoor Attack
ICLR 2026
pub
Jianwei Li, Jung-Eun Kim
Safety Alignment Can Be Not Superficial With Explicit Safety Signals
Main PaperDefense Jailbreak Attack
ICML 2025
pub
Jianwei Li, Jung-Eun Kim
Superficial Safety Alignment Hypothesis
Main PaperDefense Fine-tuning Attack
ICLR 2026
pub
Jianwei Li, Sheng Liu, Qi Lei
Beyond Gradient and Priors in Privacy Attacks: Leveraging Pooler Layer Inputs of Language Models in Federated Learning
Workshop OralDefense Privacy Leakage
FL@FM-NeurIPS 2023
pub
Jianwei Li, Yijun Dong, Qi Lei
Greedy Output Approximation: Towards Efficient Structured Pruning for LLMs Without Retraining
Main Paper
CPAL 2025
Teaching & Research Assistant
  • 2023.8-2023.12, NCSU, Teachin Assistant for CSC422-Fall 2023, Automated Learning and Data Analysis
  • 2024.1-2024.5, NCSU, Teachin Assistant for CSC422-Spring 2023, Machine Learning
  • 2024.6~present, NCSU, Reserach Assistant at Kim Lab
🎸 My Favorite Bands
刺猬
刺猬
回春丹
回春丹
麻园诗人
麻园诗人
“Music fuels creativity 🎧”
© 2022 jianwei.li All rights reserved
(Last update: Jan 29, 2026.)