Posts by Collection

Portfolio item number 2

Short description of portfolio item number 2

AIRS: Explanation for Deep Reinforcement Learning based Security Applications

Published in USENIX Security Symposium 2023, 2023

Recommended citation: Jiahao Yu, Wenbo Guo, Qi Qin, Gang Wang, Ting Wang, Xinyu Xing. (2023). "AIRS: Explanation for Deep Reinforcement Learning based Security Applications." USENIX Security.
Download Paper

StateMask: Explaining Deep Reinforcement Learning through State Mask

Published in Neural Information Processing Systems (NeurIPS) 2023, 2023

Recommended citation: Zelei Cheng*, Xian Wu*, Jiahao Yu*, Wenhai Sun, Wenbo Guo, Xinyu Xing. (2023). "StateMask: Explaining Deep Reinforcement Learning through State Mask." NeurIPS.
Download Paper

BandFuzz: A Practical Framework for Collaborative Fuzzing with Reinforcement Learning

Published in ICSE Workshop on Search-Based and Fuzz Testing (SBFT) 2024, 2024

First prize at the SBFT 2024 Fuzzing Competition.

Recommended citation: Wenxuan Shi, Hongwei Li, Jiahao Yu, Wenbo Guo, Xinyu Xing. (2024). "BandFuzz: A Practical Framework for Collaborative Fuzzing with Reinforcement Learning." ICSE@SBFT.
Download Paper

Assessing Prompt Injection Risks in 200+ Custom GPTs

Published in ICLR Workshop on Secure and Trustworthy Large Language Models 2024, 2024

Featured in WIRED.

Recommended citation: Jiahao Yu, Yuhang Wu, Dong Shu, Mingyu Jin, Xinyu Xing. (2024). "Assessing Prompt Injection Risks in 200+ Custom GPTs." ICLR@SeT-LLM.
Download Paper

RICE: Breaking Through the Training Bottlenecks of Reinforcement Learning with Explanation

Published in International Conference on Machine Learning (ICML) 2024, 2024

ICML 2024 Spotlight (top 3.5%).

Recommended citation: Zelei Cheng*, Xian Wu*, Jiahao Yu*, Sabrina Yang, Gang Wang, Xinyu Xing. (2024). "RICE: Breaking Through the Training Bottlenecks of Reinforcement Learning with Explanation." ICML Spotlight.

LLM-Fuzzer: Scaling Assessment of Large Language Model Jailbreaks

Published in USENIX Security Symposium 2024, 2024

Originally released as GPTFuzzer. Geekcon 2023 Breakthrough Award. Adopted by major LLM providers and integrated into Microsoft Azure PyRIT.

Recommended citation: Jiahao Yu, Xingwei Lin, Zheng Yu, Xinyu Xing. (2024). "LLM-Fuzzer: Scaling Assessment of Large Language Model Jailbreaks." USENIX Security.
Download Paper

Soft-Label Integration for Robust Toxicity Classification

Published in Neural Information Processing Systems (NeurIPS) 2024, 2024

Featured in MIT Technology Review China.

Recommended citation: Zelei Cheng*, Xian Wu*, Jiahao Yu*, Shuo Han, Xin-Qiang Cai, Xinyu Xing. (2024). "Soft-Label Integration for Robust Toxicity Classification." NeurIPS.
Download Paper

The Illusion of Role Separation: Hidden Shortcuts in LLM Role Learning (and How to Fix Them)

Published in International Conference on Machine Learning (ICML) 2025, 2025

Recommended citation: Zihao Wang, Yibo Jiang, Jiahao Yu, Heqing Huang. (2025). "The Illusion of Role Separation: Hidden Shortcuts in LLM Role Learning (and How to Fix Them)." ICML.

Mind the Inconspicuous: Revealing the Hidden Weakness in Aligned LLMs’ Ethical Boundaries

Published in USENIX Security Symposium 2025, 2025

USENIX Security 2025 Long Talk.

Recommended citation: Jiahao Yu, Haozheng Luo, Jerry Yao-Chieh Hu, Wenbo Guo, Yan Chen, Han Liu, Xinyu Xing. (2025). "Mind the Inconspicuous: Revealing the Hidden Weakness in Aligned LLMs’ Ethical Boundaries." USENIX Security.
Download Paper

PatchAgent: A Practical Program Repair Agent Mimicking Human Expertise

Published in USENIX Security Symposium 2025, 2025

USENIX Security 2025 Long Talk. CSAW 2025 Technical Impact Runner-up. Foundation of the AIxCC Team 42-b3yond-bug solution.

Recommended citation: Zheng Yu, Ziyi Guo, Yuhang Wu, Jiahao Yu, Meng Xu, Dongliang Mu, Yan Chen, Xinyu Xing. (2025). "PatchAgent: A Practical Program Repair Agent Mimicking Human Expertise." USENIX Security.
Download Paper

BlockScan: Detecting Anomalies in Blockchain Transactions

Published in Neural Information Processing Systems (NeurIPS) 2025, 2025

Recommended citation: Jiahao Yu, Xian Wu, Hao Liu, Wenbo Guo, Xinyu Xing. (2025). "BlockScan: Detecting Anomalies in Blockchain Transactions." NeurIPS.
Download Paper

GPO: Learning from Critical Steps to Improve LLM Reasoning

Published in Neural Information Processing Systems (NeurIPS) 2025, 2025

Featured in MIT Technology Review China.

Recommended citation: Jiahao Yu, Zelei Cheng, Xian Wu, Xinyu Xing. (2025). "GPO: Learning from Critical Steps to Improve LLM Reasoning." NeurIPS.
Download Paper

Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations

Published in International Conference on Machine Learning (ICML) 2026, 2026

Recommended citation: Haozheng Luo, Yimin Wang, Jiahao Yu, Binhui Wang, Yan Chen. (2026). "Contrastive Reasoning Alignment: Reinforcement Learning from Hidden Representations." ICML.

Decoupled Alignment for Robust Plug-and-Play Adaptation

Published in Conference on Language Modeling (COLM) 2026, 2026

Recommended citation: Haozheng Luo*, Jiahao Yu*, Wenxin Zhang, Jialong Li, Chenghao Qiu, Yimin Wang, Eric Hanchen Jiang, Jerry Yao-Chieh Hu, Yan Chen, Binghui Wang, Xinyu Xing, Han Liu. (2026). "Decoupled Alignment for Robust Plug-and-Play Adaptation." COLM.
Download Paper

Locus: Agentic Predicate Reasoning for Directed Fuzzing

Published in International Conference on Software Engineering (ICSE) 2026, 2026

Recommended citation: Jie Zhu, Chihao Shen, Ziyang Li, Jiahao Yu, Yizheng Chen, Kexin Pei. (2026). "Locus: Agentic Predicate Reasoning for Directed Fuzzing." ICSE.

PROMPTFUZZ: Harnessing Fuzzing Techniques for Robust Testing of Prompt Injection in LLMs

Published in IEEE Transactions on Information Forensics and Security (TIFS), 2026

Recommended citation: Jiahao Yu*, Yangguang Shao*, Hanwen Miao, Junzheng Shi. (2026). "PROMPTFUZZ: Harnessing Fuzzing Techniques for Robust Testing of Prompt Injection in LLMs." IEEE TIFS.
Download Paper

Building Coding Agents via Entropy-Enhanced Multi-Turn Preference Optimization

Published in arXiv preprint arXiv:2509.12434, 2026

1st on SWE-bench Lite and 5th on SWE-bench Verified among open-weight models.

Recommended citation: Jiahao Yu, Zelei Cheng, Xian Wu, Xinyu Xing. (2026). "Building Coding Agents via Entropy-Enhanced Multi-Turn Preference Optimization." arXiv:2509.12434.
Download Paper