A novel agent security software testing scheme driven by large language model

doi:10.12267/j.issn.2096-5931.2026.08.011

Abstract

Abstract:

With the large-scale deployment of autonomous AI Agents in low-altitude intelligent network, industrial control, financial services and urban governance, new security vulnerabilities such as prompt injection, unauthorized tool calling, context poisoning and multi-agent trust hijacking continue to break out. Traditional testing methods including static code scanning, dynamic payload blasting and manual red team penetration suffer from low vulnerability detection accuracy, high false positive rate, inability to cover emergent risks in long link tasks and insufficient efficiency of test case generation. A hierarchical agent security testing scheme driven by retrieval-augmented large language model is proposed to meet the full-lifecycle security verification requirements of AI Agents. The scheme can effectively identify emergent security defects in multi-step chains that cannot be captured by traditional testing solutions, providing a complete technical path for standardized security testing of agent software.

Key words: agent, security testing, large language model, retrieval-augmented generation, vulnerability detection

CLC Number:

TN929.5

HAN Han, WANG Jingyao, CHANG Tianen. A novel agent security software testing scheme driven by large language model[J]. Information and Communications Technology and Policy, 2026, 52(8): 74-81.

Add to citation manager EndNote|Ris|BibTeX

URL:

http://ictp.caict.ac.cn/EN/10.12267/j.issn.2096-5931.2026.08.011

http://ictp.caict.ac.cn/EN/Y2026/V52/I8/74

Figures/Tables 5

References 10

[1]	工业和信息化部人工智能安全委员会. AI智能体安全测试体系建设白皮书[R], 2025.
[2]	江颉, 王豪, 李明达, 等. 基于大语言模型的多智能体自动渗透测试框架构建与评估[J]. 自动化学报, 2026, 52 (4): 821-832.
[3]	李美桃, 刘永东. 生成式人工智能应用安全测评的实施路径[J]. 智能安全, 2025, 41 (1): 45-51.
[4]	Levo AI. Why legacy security testing fails AI agentic applications[R], 2025.
[5]	OpenWeb Application Security Project. OWASP ASI top 10 2026智能体安全风险标准[R], 2026.
[6]	陈永康, 李虑, 赵重阳, 等. AI辅助的智能系统安全性测试技术[J]. 智能安全, 2024, 3 (4): 72-80.
[7]	Khule T, Marksteiner S, Alguindigue J, et al, STAF:leveraging LLMs for automated attack tree-based security test generation[R], 2025.
[8]	康绍莉, 王可, 秦海超, 等. 面向6G的低空智能网联系统用例研究[J]. 信息通信技术与政策, 2026, 52 (4): 2-14. doi: 10.12267/j.issn.2096-5931.2026.04.001
[9]	Agent-SafetyBench. Comprehensive evaluation framework for LLM agent security[R], 2024.
[10]	微软安全研究院. PyRIT大模型红队测试框架技术规范[R], 2025.

测试方案	攻击面覆盖缺陷	准确率短板	误报问题	自动化适配能力
SAST	仅覆盖代码层,缺失动态交互、多智能体场景	动态漏洞漏检率>42%	低,仅规则匹配告警	完全自动化
DAST	仅单轮输入测试,无长链路、协同攻击仿真	隐式注入漏洞检出率<30%	载荷无语义上下文,误报率>40%	完全自动化
人工红队渗透	受专家知识边界限制,小众攻击场景易遗漏	综合检出率为65.15%	人工复核过滤,误报率低	无自动化能力

测试方案	攻击面覆盖缺陷	准确率短板	误报问题	自动化适配能力
SAST	仅覆盖代码层,缺失动态交互、多智能体场景	动态漏洞漏检率>42%	低,仅规则匹配告警	完全自动化
DAST	仅单轮输入测试,无长链路、协同攻击仿真	隐式注入漏洞检出率<30%	载荷无语义上下文,误报率>40%	完全自动化
人工红队渗透	受专家知识边界限制,小众攻击场景易遗漏	综合检出率为65.15%	人工复核过滤,误报率低	无自动化能力

测试方案	检出准确率	漏检率	召回率	单套测试耗时/h
SAST+DAST	56.82%	42.79%	57.21%	1.2
人工红队渗透	90.45%	34.85%	65.15%	12.6
PentestGPT基础框架	71.36%	21.63%	78.37%	3.7
单层大模型测试	78.51%	14.22%	85.78%	2.4
分层RAG大模型方案	96.37%	3.63%	96.37%	2.6

测试方案	检出准确率	漏检率	召回率	单套测试耗时/h
SAST+DAST	56.82%	42.79%	57.21%	1.2
人工红队渗透	90.45%	34.85%	65.15%	12.6
PentestGPT基础框架	71.36%	21.63%	78.37%	3.7
单层大模型测试	78.51%	14.22%	85.78%	2.4
分层RAG大模型方案	96.37%	3.63%	96.37%	2.6

[1]	LIU Jun, DONG Ji, WANG Wei. Research on security protection and governance paths of terminal agents from perspective of risk management and control [J]. Information and Communications Technology and Policy, 2026, 52(8): 18-22.
[2]	WU Hanbing, FENG Yizhuo. Research on user authorization mechanisms and identity forgery risks in agent interconnection scenarios [J]. Information and Communications Technology and Policy, 2026, 52(8): 2-10.
[3]	XU Ming. A review of cross-domain data security governance for low-altitude one-network unified flight platform [J]. Information and Communications Technology and Policy, 2026, 52(6): 70-78.
[4]	TIAN Xiaoyu, LI Wenyu, BI Chunli, FU Na, ZHANG Leilei. A review of research on Chinese datasets for ethical evaluation of large language models [J]. Information and Communications Technology and Policy, 2026, 52(5): 58-68.
[5]	WANG Shiji, WANG Liwen, ZHANG Yan, CAO Chang, WANG Runyan, HE Wanxian, PEI Pei. Service semantic routing technology for application-level computing power services [J]. Information and Communications Technology and Policy, 2026, 52(5): 80-88.
[6]	CHENG Ying, LIU Zhipeng, ZHANG Meng. Research on development pathways, potential impacts, and governance strategies of terminal intelligent agents [J]. Information and Communications Technology and Policy, 2026, 52(3): 2-7.
[7]	YANG Lei, WANG Feng, ZHANG Xin, LIN Jun. The ecological game and comparison of cooperative modes between intelligent terminals and internet services in the large model era [J]. Information and Communications Technology and Policy, 2026, 52(3): 8-13.
[8]	NING Keyu, MA Fei, LI Zhe, DONG Xiaohui. A review of post-training cost optimization technology for large language models based on distributed computing optimization [J]. Information and Communications Technology and Policy, 2026, 52(2): 44-52.
[9]	WEI Tiancheng, GUO Zhen, YANG Yunlong. Research on the paths and strategies of empowering smart city construction with large language models [J]. Information and Communications Technology and Policy, 2025, 51(8): 91-96.
[10]	GE Jian, NIU Xiaoyan, BI Ran, HUANG Yongtao. Low-carbon AI: research on green training and inference optimization methods for large model [J]. Information and Communications Technology and Policy, 2025, 51(6): 44-51.
[11]	LIU Gaofeng, JIANG Yuehan, ZHANG Yuxiong. Artificial intelligence enables high-level development and high-quality construction of digital twin city [J]. Information and Communications Technology and Policy, 2025, 51(3): 74-82.
[12]	LIU Dongfang, YANG Tiankai, CHANG Zheng, HAO Pengfei. Mechanism innovation and practical exploration of large language models empowering government investment project evaluation [J]. Information and Communications Technology and Policy, 2025, 51(12): 48-56.
[13]	XIN Xing, WU Jun, WANG Haiyan, ZHANG Yang, GUO Wei. Research on the application scenarios and trends of AI in empowering telecom user complaint management [J]. Information and Communications Technology and Policy, 2025, 51(12): 57-62.
[14]	HUANG Fengyu, HUANG Yunxia, FU Rongrong, ZHAO Qiannan. Research on the development status of artificial intelligence terminal industry [J]. Information and Communications Technology and Policy, 2025, 51(12): 92-96.
[15]	WANG Wei, PANG Xiaobo. Research on application strategies for government large models [J]. Information and Communications Technology and Policy, 2025, 51(11): 74-80.