About

Xingwei(Roy) Lin is a Security Algorithm Expert at Ant Group, focusing on the intersection of AI and Security: securing AI systems (e.g., LLM safety and robustness) and applying LLM-powered agents to software & network security — from vulnerability detection to automated patching. He has discovered over 100 vulnerabilities in widely-deployed software such as Apple, Microsoft, and QEMU, earning Pwnie Awards "Most Innovative Research" nominations in both 2020 and 2021. His work has been published extensively in top-tier venues and received the ACM CCS 2021 Best Paper Award.

He is also pursuing a PhD at Zhejiang University under the supervision of Prof. Chunming Wu. He received his M.S. from UESTC under the supervision of Prof. Xiaosong Zhang and Prof. Ting Chen, and was a Research Assistant at PolyU working with Prof. Xiapu Luo.

Research Interests

  • AI Security — LLM/Agent safety, robustness, and jailbreak assessment
  • LLM-powered Security Agents — Vulnerability detection and automated patch generation
  • System & Software Security — Fuzzing, program analysis and binary vulnerability discovery

News

  • [2026.06] Our paper ChainDelta, an LLM-driven fuzzing-agent framework for automatic patch-based exploit generation on Ethereum, is accepted to FSE 2026.
  • [2026.05] Our paper Code Language Models for Security Patch Management: How Far Are We?, the first large-scale empirical study of nine CodeLMs on three patch management tasks, is accepted to IEEE Transactions on Services Computing (TSC).
  • [2026.01] Our paper ICON, an efficient multi-turn jailbreak attack framework leveraging intent-context coupling against LLMs, is now available on arXiv.
  • [2025] Our paper Hyperion, an LLM and dataflow-guided symbolic execution framework for unveiling DApp inconsistencies, is accepted at ICSE 2025.
  • [2025] Our paper AePPollo on LLM-powered prototype pollution vulnerability detection and exploit generation is accepted at SEKE 2025.

Experience

2019.07 – Present Ant Group, Security Algorithm Expert
2017.07 – 2017.12 The Hong Kong Polytechnic University, Research Assistant

Education

2024.09 – Present Zhejiang University, Ph.D. in Computer Science and Technology
2016.09 – 2019.06 University of Electronic Science and Technology of China, M.S. in Computer Science and Technology
2012.09 – 2016.07 University of Electronic Science and Technology of China, B.E. in Communication Engineering

Selected Publications

* corresponding author. Full list: Google Scholar.

  • FSE'26 (CCF-A) ChainDelta: Automatic Patch-Based Exploit Generation for Ethereum with Fuzzing Agents [pdf]
    Mingxi Ye, Yuhong Nan, Zhijie Zhong, Jianzhong Su, Xingwei Lin, Peilin Zheng, Zibin Zheng
  • TSC'26 (CCF-A) Code Language Models for Security Patch Management: How Far Are We? [paper]
    Xingwei Lin, Sicong Cao, Le Yu, Xiaobing Sun, Fu Xiao, Lei Xue, Chunming Wu, Kui Ren, David Lo
  • arXiv'26 ICON: Intent-Context Coupling for Efficient Multi-Turn Jailbreak Attack [pdf]
    Xingwei Lin, Wenhao Lin, Sicong Cao, Jiahao Yu, Renke Huang, Lei Xue, Chunming Wu
  • ICSE'25 (CCF-A) Hyperion: Unveiling DApp Inconsistencies Using LLM and Dataflow-Guided Symbolic Execution [pdf]
    Shuo Yang, Xingwei Lin, Jiachi Chen, Qingyuan Zhong, Lei Xiao, Renke Huang, Yanlin Wang, Zibin Zheng
  • TOSEM'25 (CCF-A) Scuzer: A Scheduling Optimization Fuzzer for TVM [pdf]
    Xiangxiang Chen, Xingwei Lin, Jingyi Wang, Jun Sun, Jiashui Wang, Wenhai Wang
  • SEKE'25 (CCF-C) AePPollo: Automated Exploit Generation for Prototype Pollution Vulnerabilities in Node.js Application [pdf]
    Wenya Wang, Wang Gao, Zhenghe Wang, Sicong Cao, Xingwei Lin*, Dawu Gu
  • USENIX Security'24 (CCF-A) LLM-Fuzzer: Scaling Assessment of Large Language Model Jailbreaks [pdf]
    Jiahao Yu, Xingwei Lin, Zheng Yu, Xinyu Xing
  • USENIX Security'24 (CCF-A) Code is not Natural Language: Unlock the Power of Semantics-Oriented Graph Representation for Binary Code Similarity Detection [pdf]
    Haojie He, Xingwei Lin, Ziang Weng, Ruijie Zhao, Shuitao Gan, Libo Chen, Yuede Ji, Jiashui Wang, Zhi Xue
  • CCS'23 (CCF-A) HODOR: Shrinking Attack Surface on Node.js via System Call Limitation [pdf]
    Wenya Wang, Xingwei Lin, Jingyi Wang, Wang Gao, Dawu Gu, Wei Lv, Jiashui Wang
  • CCS'21 (CCF-A) Best Paper V-SHUTTLE: Scalable and Semantics-Aware Hypervisor Virtual Device Fuzzing [pdf]
    Gaoning Pan, Xingwei Lin, Xuhong Zhang, Yongkang Jia, Shouling Ji, Chunming Wu, Xinlei Ying, Jiashui Wang, Yanjun Wu
  • USENIX Security'21 (CCF-A) APICRAFT: Fuzz Driver Generation for Closed-source SDK Libraries [pdf]
    Cen Zhang, Xingwei Lin, Yuekang Li, Yinxing Xue, Jundong Xie, Hongxu Chen, Xinlei Ying, Jiashui Wang, Yang Liu
  • CCS'18 (CCF-A) JN-SAF: Precise and Efficient NDK/JNI-aware Inter-language Static Analysis Framework for Security Vetting of Android Applications with Native Code [pdf]
    Fengguo Wei, Xingwei Lin, Xinming Ou, Ting Chen, Xiaosong Zhang
  • Black Hat Europe'23 HODOR: Reducing Attack Surface on Node.js via System Call Limitation [talk]
    Wenya Wang, Xingwei Lin, Jingyi Wang, Wang Gao, Dawu Gu
  • Black Hat Asia'21 Scavenger: Misuse Error Handling Leading to Qemu/KVM Escape [talk]
    Gaoning Pan, Xingwei Lin, Xinlei Ying, Jiashui Wang, Chunming Wu

Awards & Honors

  • 2023 Frontier Breakthrough Award & Outstanding Presentation Award, GeekCon
  • 2023 First Prize, 3rd Trusted Blockchain Security Attack-Defense Competition
  • 2023 Best Paper Award, BlockSys 2023
  • 2021 Best Paper Award, ACM CCS 2021 (V-SHUTTLE)
  • 2021 Pwnie Awards Nomination, "Most Innovative Research" (V-SHUTTLE)
  • 2021 2nd Place, Baidu BCTF AutoPwn
  • 2020 Pwnie Awards Nomination, "Most Innovative Research" (APICRAFT)

Professional Service

  • Program Committee — ACM CCS 2024 (Software Security Track)
  • Program Committee — ACM AsiaCCS 2027 (AI Security Track)
  • Journal Reviewer — Automated Software Engineering (ASE), 2026

CVE Contributions

Responsibly disclosed 90+ CVEs across major software vendors.

Apple (75 CVEs)
CVE-2020-9790, CVE-2020-9871, CVE-2020-9872, CVE-2020-9873, CVE-2020-9874, CVE-2020-9877, CVE-2020-9879, CVE-2020-9888, CVE-2020-9889, CVE-2020-9890, CVE-2020-9891, CVE-2020-9937, CVE-2020-9938, CVE-2020-9955, CVE-2020-9960, CVE-2020-9961, CVE-2020-9980, CVE-2020-11758, CVE-2020-11759, CVE-2020-11760, CVE-2020-11761, CVE-2020-11762, CVE-2020-11763, CVE-2020-11764, CVE-2020-11765, CVE-2020-27908, CVE-2020-27909, CVE-2020-27910, CVE-2020-27912, CVE-2020-27919, CVE-2020-27927, CVE-2020-27933, CVE-2020-27939, CVE-2020-29608, CVE-2020-29615, CVE-2020-29617, CVE-2020-29618, CVE-2020-29619, CVE-2020-29625, CVE-2021-1736, CVE-2021-1737, CVE-2021-1741, CVE-2021-1742, CVE-2021-1743, CVE-2021-1744, CVE-2021-1746, CVE-2021-1754, CVE-2021-1773, CVE-2021-1774, CVE-2021-1777, CVE-2021-1778, CVE-2021-1783, CVE-2021-1785, CVE-2021-1793, CVE-2021-1811, CVE-2021-1814, CVE-2021-1818, CVE-2021-1880, CVE-2021-1881, CVE-2021-30743, CVE-2021-30753, CVE-2021-30755, CVE-2021-30831, CVE-2021-30840, CVE-2021-30841, CVE-2021-30842, CVE-2021-30843, CVE-2021-30911, CVE-2021-30929, CVE-2021-30939, CVE-2021-30940, CVE-2021-30941, CVE-2022-42810, CVE-2020-36521, CVE-2025-24127
Oracle VirtualBox (9 CVEs)
CVE-2020-2905, CVE-2020-14872, CVE-2020-14889, CVE-2020-14892, CVE-2021-2073, CVE-2021-2074, CVE-2021-2086, CVE-2021-2123, CVE-2021-2130
QEMU (2 CVEs)
CVE-2020-14394, CVE-2020-27661
Microsoft (1 CVE)
CVE-2020-1492
SQLite (1 CVE)
CVE-2019-16168
OpenEXR (1 CVE)
CVE-2020-15306
libextractor (3 CVEs)
CVE-2018-14346, CVE-2018-14347, CVE-2018-16430