摘要： 现有浏览器JavaScript引擎Fuzz工具在检测非崩溃类漏洞时存在不足。为此，提出一种结合内存检测与大语言模型的增强型漏洞检测方法。该方法改进Fuzz工具对潜在漏洞的判别能力，降低漏报率。同时，将大语言模型与常规变异器相结合，在PoC集激励下生成更复杂且具针对性的测试用例，有效探索深层代码路径，提升代码覆盖率。实验结果显示，改进后的工具在分支覆盖率和漏洞检测能力上较现有工具有效提升。

中图分类号：TP309 文献标志码：A DOI: 10.16157/j.issn.0258-7998.256463
中文引用格式： 丁加宇，沙乐天，潘家晔. 基于大语言模型的JS引擎模糊测试方法[J]. 电子技术应用，2026，52(2)：66-70.
英文引用格式： Ding Jiayu，Sha Letian，Pan Jiaye. Fuzz testing method for JavaScript engines based on large language model[J]. Application of Electronic Technique，2026，52(2)：66-70.

Fuzz testing method for JavaScript engines based on large language model

Abstract： Existing Fuzz tools for browser JavaScript engines are insufficient in detecting non-crashing vulnerabilities. To address this issue, an enhanced vulnerability detection method combining memory detection and large language models (LLMs) is proposed. This method improves the Fuzz tool's ability to identify potential vulnerabilities and reduces the false-negative rate. Additionally, the integration of LLMs with conventional mutators generates more complex and targeted test cases under the stimulus of Proof-of-Concept (PoC) sets, effectively exploring deeper code paths and increasing code coverage. Experimental results show that the improved tool significantly enhances branch coverage and vulnerability detection capabilities compared to existing tools.

Key words : Fuzz testing；JavaScript engine；use after free