中图分类号 : TP309. 2 文献标志码 : A DOI :10.19358/j.issn.2097-1788.2026.02.003 中文引用格式 : 覃锦端 , 尉雯雯 , 王月兵 , 等. 基于大模型的深层 Web 越权漏洞检测方法 [J]. 网络安全与数据治理 , 2026 , 45 (2) : 20 -27. 英文引用格式 : Qin Jinduan, Wei Wenwen, Wang Yuebing, et al. A deep Web privilege escalation vulnerability detection method based on large models [J]. Cyber Security and Data Governance, 2026 , 45(2) : 20 -27.
A deep Web privilege escalation vulnerability detection method based on large models
Qin Jinduan1 , Wei Wenwen2 , Wang Yuebing1 , Liu Zunliang1 , Liu Cong1
1. Hangzhou Meichuang Technology Co. , Ltd. ; 2. The First Affiliated Hospital (Southwest Hospital) of Third Military Medical University (Army Medical University)
Abstract： Privilege escalation vulnerability detection and mining is an important topic in traditional Web application security. Due to its wide coverage, deep concealment, and lack of fixed traffic characteristics, privilege escalation vulnerabilities have always been a difficult point in the governance of Web application vulnerabilities. Currently, the common Web privilege escalation vulnerability mining methods in the industry mainly rely on passive detection plugins combined with manual mining. The principle of the current passive detection plugins is mostly to re- place with high-privilege or same-privilege account credentials, and then use the length of the returned traffic packet as the basis for determi- ning whether there is a vulnerability. Although this method can save some manual testing costs, it can only detect shallow privilege escalation vulnerabilities, and for parameter-level privilege escalation vulnerabilities, it still relies on manual testing. Based on large model technology, this paper proposes a passive deep Web privilege escalation vulnerability detection method, aiming to automatically identify the meaning of pa- rameter names and the characteristics of parameter values through large models, dynamically generate test parameters for sending, and use mul- tiple dimensions such as the length and specific content of the returned traffic packet as the basis for vulnerability judgment. After testing, this method can detect deeper privilege escalation vulnerabilities and effectively save manual mining costs.
Key words : Web vulnerability; vulnerability detection; privilege escalation vulnerability; large model
