中圖分類號：TP309文獻標志碼：ADOI:10.19358/j.issn.2097-1788.2026.07.003中文引用格式：李金龍，趙新元.基于LLM的前端加密自動分析框架［J］.網絡安全與數據治理，2026，45（7）：16-23.
英文引用格式：Li Jinlong，Zhao Xinyuan. LLMbased automated analysis framework for frontend encryption［J］.Cyber Security and Data Governance，2026，45（7）：16-23.

LLM-based automated analysis framework for front-end encryption

Abstract： To raise the barrier to cyberattacks and mitigate contentbased threats, many websites use frontend scripts to encrypt data packets. However, the transparency of frontend scripts also exposes their encryption logic and keys to the potential risk of reverse engineering. Currently, the identification and analysis of such risks primarily rely on manual reverse engineering by security experts, which is inefficient, highly dependent on expert experience, and unable to meet the demands of largescale security assessments. To address this issue, we propose an automated frontend encryption analysis framework based on large language models (LLMs). This framework uses dynamic hooking techniques to capture calls to encryption functions and constructs prompts specifically designed for encryption reverse engineering to drive the LLM to intelligently parse obfuscated code. This ensures that the framework can automatically reconstruct complete encryption and decryption wrapper functions. Additionally, we introduce verifiability constraints to ensure the reliability of the analysis results. Experimental results on six realworld, anonymized university identity authentication platforms demonstrate that the framework successfully captures cryptographic details across various scenarios, including those involving dynamic key acquisition, and effectively improves the efficiency of such security assessments. This research provides a scalable technical approach for achieving efficient, largescale security assessments of frontend encryption mechanisms.

Key words : front-end encryption; LLM; dynamic hooks; automation; web security