TY - GEN
T1 - Bridging Expert Reasoning and LLM Detection
T2 - 35th ACM Web Conference, WWW 2026
AU - Guo, Wenbo
AU - Song, Shiwen
AU - Guo, Jiaxun
AU - Xu, Zhengzi
AU - Liu, Chengwei
AU - Ou, Haoran
AU - Ge, Mengmeng
AU - Liu, Yang
N1 - Publisher Copyright:
© 2026 Owner/Author.
PY - 2026/4/12
Y1 - 2026/4/12
N2 - Open-source ecosystems such as NPM and PyPI are increasingly targeted by supply chain attacks, yet existing detection methods either depend on fragile handcrafted rules or data-driven features that fail to capture evolving attack semantics. We present IntelGuard, a retrieval-augmented generation (RAG) based framework that integrates expert analytical reasoning into automated malicious package detection. IntelGuard constructs a structured knowledge base from over 8,000 threat intelligence reports, linking malicious code snippets with behavioral descriptions and expert reasoning. When analyzing new packages, it retrieves semantically similar malicious examples and applies LLM-guided reasoning to assess whether code behaviors align with intended functionality. Experiments on 4,027 real-world packages show that IntelGuard achieves 99% accuracy and a 0.50% false positive rate, while maintaining 96.5% accuracy on obfuscated code. Deployed on PyPI.org, it discovered 54 previously unreported malicious packages, demonstrating interpretable and robust detection guided by expert knowledge.
AB - Open-source ecosystems such as NPM and PyPI are increasingly targeted by supply chain attacks, yet existing detection methods either depend on fragile handcrafted rules or data-driven features that fail to capture evolving attack semantics. We present IntelGuard, a retrieval-augmented generation (RAG) based framework that integrates expert analytical reasoning into automated malicious package detection. IntelGuard constructs a structured knowledge base from over 8,000 threat intelligence reports, linking malicious code snippets with behavioral descriptions and expert reasoning. When analyzing new packages, it retrieves semantically similar malicious examples and applies LLM-guided reasoning to assess whether code behaviors align with intended functionality. Experiments on 4,027 real-world packages show that IntelGuard achieves 99% accuracy and a 0.50% false positive rate, while maintaining 96.5% accuracy on obfuscated code. Deployed on PyPI.org, it discovered 54 previously unreported malicious packages, demonstrating interpretable and robust detection guided by expert knowledge.
KW - large language models
KW - open-source ecosystems
KW - retrieval-augmented generation
KW - supply chain security
KW - threat intelligence
UR - https://www.scopus.com/pages/publications/105038562660
U2 - 10.1145/3774904.3792083
DO - 10.1145/3774904.3792083
M3 - 会议稿件
AN - SCOPUS:105038562660
T3 - WWW 2026 - Proceedings of the ACM Web Conference 2026
SP - 3554
EP - 3565
BT - WWW 2026 - Proceedings of the ACM Web Conference 2026
PB - Association for Computing Machinery, Inc
Y2 - 29 June 2026 through 3 July 2026
ER -