AI Safety Researcher

Hi, I'm Shihao Weng 翁诗浩.

I build tools that make AI agents more trustworthy.

I am currently a Ph.D. student at Nanjing University with Yang Feng, visiting SMU with Xiaofei Xie.

Papers Email
now In Singapore · Feel free to reach out to discuss anything related to AI.

Research

Currently working on two directions:

01

Self-evolving Agents

An agent that can improve itself toward more trustworthy behavior over time.

02

Agent Defense

Defending AI agents against adversarial input and compromised tools.

News

  1. 2025.12 Two papers accepted at ICSE'26 (Cycle 2) CCF-A
  2. 2025.06 One paper accepted at ICSE'26 CCF-A
  3. 2024.09 One co-authored paper accepted at ISSTA'24 CCF-A
  4. 2023.11 One co-authored paper accepted at FSE'23 CCF-A
  5. 2023.10 Awarded the Distinguished Paper Award at Internetware'23 CCF-C

Publications

* indicates corresponding author.

Highlights

ARGUS teaser figure
Preprint 2026

ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection

We propose a provenance-based defense that traces how untrusted context propagates into an agent's decisions and blocks any action not backed by trustworthy evidence.

Shihao Weng, Yang Feng*, Jinrui Zhang, Xiaofei Xie, Jiongchi Yu, Jia Liu.

↗ paper {} code

All publications

2026
Preprint

Beyond Accuracy: Policy Invariance as a Reliability Test for LLM Safety Judges

Shihao Weng, Yang Feng*, Xiaofei Xie.

↗ paper {} code
Preprint

ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection

Shihao Weng, Yang Feng*, Jinrui Zhang, Xiaofei Xie, Jiongchi Yu, Jia Liu.

↗ paper {} code
ICSE'26 CCF-A

TACO: Trust Assessment of Large Language Models in Coding Assistance Tasks

Shihao Weng, Yang Feng*, Jincheng Li, Yining Yin, Zhenlun Zhang, Lyuxi Liu, Jia Liu.

⚙︎ tool
ICSE'26 CCF-A

VADA: A Multicultural Benchmark for Value-Aware Data Generation and Alignment Evaluation in LLMs

Zhenlun Zhang, Yang Feng*, Shihao Weng, Yining Yin, Jincheng Li, Jia Liu.

ICSE'26 CCF-A

AtPatch: Debugging Transformers via Hot-Fixing Over-Attention

Shihao Weng, Yang Feng*, Jincheng Li, Yining Yin, Xiaofei Xie, Jia Liu.

↗ paper {} code
2025
FCS'25 CCF-B

Data preparation and quality for code-centric generative software engineering tasks: a systematic literature review

Shihao Weng, Yang Feng*, Yining Yin, Zhenlun Zhang, Baowen Xu.

↗ paper
Internetware'25 CCF-C

Lightweight Probabilistic Coverage Metrics for Efficient Testing of Deep Neural Networks

Yining Yin, Yang Feng*, Shihao Weng, Xinyu Gao, Jia Liu, Zhihong Zhao.

↗ paper {} code
2024
EMSE'24 CCF-B

Seeing the invisible: test prioritization for object detection system

Shihao Weng, Yang Feng*, Yining Yin, Yuxuan Dai, Jia Liu, Zhihong Zhao.

↗ paper {} code
ISSTA'24 CCF-A

Datactive: Data Fault Localization for Object Detection Systems

Yining Yin, Yang Feng*, Shihao Weng, Yuan Yao, Jia Liu, Zhihong Zhao.

↗ paper {} code
2023
FSE'23 CCF-A

Dynamic Data Fault Localization for Deep Neural Networks

Yining Yin, Yang Feng*, Shihao Weng, Zixi Liu, Yuan Yao, Yichi Zhang, Zhihong Zhao, Zhenyu Chen.

↗ paper {} code
Internetware'23 CCF-C 🏆 Distinguished Paper

Prioritizing Testing Instances to Enhance the Robustness of Object Detection Systems

Shihao Weng, Yang Feng*, Yining Yin, Jia Liu.

↗ paper {} code

Honors & Awards

  • 2023–25 Outstanding Graduate Student of Nanjing University ×3
  • 2023 National Scholarship for Graduate Students
  • 2021 CCF Outstanding Undergraduate Student
  • 2020–21 National Scholarship for Undergraduate Students ×2

Talks · Service · Teaching

Coming soon

Education

  1. 2024.09 to now
    Nanjing University Ph.D. · Software Engineering
  2. 2022.09 – 2024.06
    Nanjing University M.S. · Software Engineering
  3. 2018.09 – 2022.06
    Jiangnan University B.E. · Computer Science · advised by Heng-yang Lu

Visitors