1.Guizhou Big Data Academy,Guizhou University,Guiyang 550025,Guizhou,China
2.School of Cyber Science and Engineering,Wuhan University,Wuhan 430072,Hubei,China
3.College of Information Management,Nanjing Agricultural University,Nanjing 211800,Jiangsu,China
Show less
文章历史+
Received
Published
2024-07-26
2026-02-24
Issue Date
2026-07-23
PDF (2609K)
摘要
针对高级可持续威胁(Advanced Persistent Threat,APT)溯源未考虑ATT&CK(Adversarial Tactics, Techniques, and Common Knowledge)技战术和攻击语义行为增强,未融合动静态两个视角探索和实现攻击行为互补的溯源分析,易被加壳和混淆的APT恶意软件逃避问题,提出一种基于动静态语义行为增强的APT攻击溯源(Advanced Persistent Threat Eye, APTEye)模型。首先,构建APT组织恶意软件样本集并实施预处理;其次,提取恶意软件的静态行为特征与动态行为特征;再次,设计行为特征语义增强及表征算法,分别利用Attack2Vec将静态API特征和攻击链以及语义行为映射,APISeq2Vec增强动态API序列的时间语义关系,实现低级别行为特征到高级别攻击模式的映射;接着,构建动静态特征对齐和行为语义聚合算法将APT攻击恶意软件的动态静态特征融合;最后,构建图注意力网络模型溯源APT组织。实验结果表明,APTEye模型能有效追踪溯源APT攻击,其精确率、召回率和F1值分别为92.24%、91.85%和92.04%,均优于现有模型。此外,APTEye模型能够有效识别细粒度的动静态API函数及攻击行为,实现与ATT&CK技战术映射,为后续APT攻击的意图推理和攻击阻断提供支撑。
Abstract
To address the limitations of current Advanced Persistent Threat(APT) attacks tacing methods, such as the neglect of ATT&CK(Adversarial Tactics, Techniques, and Common Knowledge) and semantic behavior enhancement, the failure to combine dynamic and static perspectives for in-depth attack traceability analysis, and the vulnerability to evasion by the packed and obfuscated APT malware. We propose a novel traceability model for APT attacks based on dynamic and static semantic behavior enhancement (APTEye). Firstly, an APT malware sample set is constructed and preprocessed. Secondly, static and dynamic behavioral features of the malware are extracted. Then, a behavioral feature semantic enhancement and representation algorithm is designed. Attack2Vec is used to map static API features, the attack chain, and the semantic behaviors. APISeq2Vec is employed to enhance the temporal semantic relationships of dynamic API sequences, thereby mapping low-level behavior features to high-level attack patterns. Subsequently, a dynamic-static feature alignment and behavior semantic aggregation algorithm is designed to integrate dynamic and static features of APT malware. Finally, a graph attention network-based model is constructed to trace APT organizations. Experimental results show that the proposed method can effectively track APT attacks, with a precision of 92.24%, a recall of 91.85%, and an F1-score of 92.04%, superior to existing models. In addition, the APTEye model can effectively identify fine-grained dynamic and static API functions and attack behaviors, facilitating the mapping between behaviors and the ATT&CK framework, which supports further intent inference and blocking of APT attacks.
APT攻击智能溯源旨在结合APT攻击的技战术特点提取溯源特征,并构建智能模型以识别攻击活动中的线索和痕迹,实现APT组织或黑客的溯源追踪。目前,学术界和产业界主要利用机器学习、深度学习、溯源图和知识图谱等方法进行APT智能溯源。文献[7-10]通过构建机器学习模型(如支持向量机、随机森林、决策树)学习攻击行为特征溯源APT组织或攻击者;文献[11-14]通过构建深度学习模型(如卷积神经网络、循环神经网络、图卷积神经网络)学习攻击审计日志或行为特征追踪APT组织;文献[15-18]利用溯源图和知识图谱关联APT攻击特征,并追踪攻击来源。然而,上述工作仍未较好地解决APT攻击溯源问题,其局限性表现为:1) 缺乏对真实世界APT攻击实战性的考虑,未能充分利用ATT&CK(Adversarial Tactics, Techniques, and Common Knowledge)框架映射和行为时序依赖关系进行APT组织溯源;2) 部分方法过度依赖专家经验和先验知识,企业倾向于利用规则匹配和同源分析来实现APT攻击溯源,缺乏细粒度和自动化的行为分析;3) 未能有效整合APT攻击的动态和静态行为特征进行语义增强,未能从动静两个视角探索攻击行为的互补作用;4) 溯源模型存在局限性,难以对抗加壳混淆技术,且无法学习适用于动静态特征的嵌入向量,导致现有工作更倾向于简单叠加的深度学习模型,与安全问题结合较浅,忽视溯源的本质且鲁棒性较差。因此,本文的研究重点在于如何结合提取的行为特征,构建融合语义知识的模型,以溯源追踪APT攻击的来源,并深入分析攻击行为。
不同APT组织所利用的攻击手法、战术行为、技术行为和关键API函数能够以结构化知识的形式呈现,并且能与各APT组织的ATT&CK框架有效映射,技战术行为与关键API函数呈现一对多的关系。表4呈现了四个代表性APT组织最常用的6种攻击行为类型,这些技战术及API函数常被应用于该组织的APT攻击事件及定向渗透活动中,它们准确反映了该APT组织的特点,通过关键特征的识别能更好地被图神经网络学习,从而追踪溯源APT组织。譬如,表4中组织2的战术行为“Discovery”和技术行为“System Information Discovery”对应于ATT&CK框架的“T1082”,通过调用“GetComputerName”和“kernel32.GetProcAddress”函数可执行“发现(信息系统)”的行为,说明该组织擅长捕获系统信息;再如组织3的战术行为“Defense Evasion”和技术行为“Obfuscated Files or Information”对应于ATT&CK框架的“T1027”,通过调用“CryptGenRandom”和“advapi32.CryptHashData”函数可加密恶意代码,并执行逃逸载荷或命令,从而躲避杀毒软件的查杀。
此外,图5清晰地展现了API函数与各类攻击行为之间的语义映射关系,并呈现出一对多的映射模式,即同一类攻击技战术往往涉及多个API函数,而某些API函数可能同时参与不同的攻击技战术。譬如,组织8通常利用ATT&CK框架中的“T1027”技术编号执行高隐蔽攻击,其战术行为为“Defense Evasion”,依赖“Obfuscated Files or Information”技术,通过文件混淆手段来实施防御逃逸和躲避系统检测,并对应关键API函数“kernel32.VirtualAllocEx”和“kernel32.VirtualProtect”,分别用于分配内存空间和修改内存保护属性。为更直观地反映其语义关系,本文在上述关系的基础上增加了5条红色辅助线,图中的辅助线表示了组织8中ATT&CK序号“T1027”、战术行为为“Defense Evasion”、“Obfuscated Files or Information”技术、API函数“kernel32.VirtualAllocEx”和“kernel32.VirtualProtect”的关联关系。由于该映射关系在该组织的多个恶意软件样本中均被观测到,因此可以推断这些API函数的使用与“Defense Evasion”战术高度相关,且不同APT组织可能通过相似的API调用序列实现攻击行为映射,更好地被后续模型学习,最终APTEye获得最佳的溯源效果。
YANGX Z, PENGG J, LIZ C, et al. Research on entity recognition and alignment of APT attack based on Bert and BiLSTM-CRF[J]. Journal on Communications, 2022, 43(6): 58-70. DOI: 10.11959/j.issn.1000-436x.2022116(Ch ).
[4]
STOJANOVIĆB, HOFER-SCHMITZK, KLEBU. APT datasets and attack modeling for automated detection methods: A review[J]. Computers & Security, 2020, 92: 101734. DOI: 10.1016/j.cose.2020.101734 .
[5]
XIONGC L, ZHUT T, DONGW H, et al. Conan: A practical real-time APT detection system with high accuracy and efficiency[J]. IEEE Transactions on Dependable and Secure Computing, 2022, 19(1): 551-565. DOI: 10.1109/TDSC.2020.2971484 .
HUANGK Z, LIANY F, FENGD G, et al. Method of cyber attack attribution based on graph model[J]. Journal of Software, 2022, 33(2): 683-698. DOI: 10.13328/j.cnki.jos.006314(Ch ).
[10]
WANGQ Q, YANH B, HANZ H. Explainable APT attribution for malware using NLP techniques[C]//Proceedings of the 2021 IEEE 21st International Conference on Software Quality, Reliability and Security (QRS). New York: IEEE Press, 2021: 70-80. DOI: 10.1109/QRS54544.2021.00018 .
[11]
CALISKANA, YAMAGUCHIF, DAUBERE, et al. When coding style survives compilation: De-anonymizing programmers from executable binaries[C]//Proceedings of the 2018 Network and Distributed System Security Symposium. Reston: Internet Society, 2018. DOI: 10.14722/ndss.2018.23304 .
[12]
HONGK F, CHENC C, CHIUY T, et al. Ctracer: Uncover C&C in advanced persistent threats based on scalable framework for enterprise log data[C]//Proceedings of the 2015 IEEE International Congress on Big Data. New York: IEEE Press, 2015: 551-558. DOI: 10.1109/BigDataCongress.2015.86 .
LIA. Research on features extraction method of attack group based on malicious code gene[D]. Beijing: Beijing University of Posts and Telecommunications, 2021. DOI: 10.26969/d.cnki.gbydu.2021.001346(Ch ).
[15]
ALRABAEES, KARBABE B, WANGL Y, et al. BinEye: Towards efficient binary authorship characterization using Deep Learning[C]//Proceedings of the 24th European Symposium on Research in Computer Security (ESORICS). Cham: Springer International Publishing, 2019: 47-67. DOI: 10.1007/978-3-030-29962-0_3 .
[16]
ROSENBERGI, SICARDG, DAVIDE. DeepAPT: Nation-state APT attribution using end-to-end deep neural networks[C]//Proceedings of 26th International Conference on Artificial Neural Networks (ICANN). Cham: Springer International Publishing, 2017: 91-99. DOI:org/10.1007/978-3-319-68612-7_11 .
[17]
ROSENBERGI, SICARDG, DAVIDE O. End-to-end deep neural networks and transfer learning for automatic analysis of nation-state malware[J]. Entropy, 2018, 20(5): 390. DOI: 10.3390/e20050390 .
[18]
HANW J, XUEJ F, WANGY, et al. APTMalInsight: Identify and cognize APT malware based on system call information and ontology knowledge framework[J]. Information Sciences, 2021, 546: 633-664. DOI: 10.1016/j.ins.2020.08.095 .
[19]
MAS Q, ZHANGX Y, XUD Y. ProTracer: Towards practical provenance tracing by alternating between logging and tainting[C]//Proceedings of the 2016 Network and Distributed System Security Symposium. Reston: Internet Society, 2016:1-15. DOI: 10.14722/ndss.2016.23350 .
[20]
BERADYA, JAUMEM, TONGV V T, et al. From TTP to IoC: Advanced persistent graphs for threat hunting[J]. IEEE Transactions on Network and Service Management, 2021, 18(2): 1321-1333. DOI: 10.1109/TNSM.2021.3056999 .
[21]
IRSHADH, CIOCARLIEG, GEHANIA, et al. TRACE: Enterprise-wide provenance tracking for real-time APT detection[J]. IEEE Transactions on Information Forensics and Security, 2021, 16: 4363-4376. DOI: 10.1109/TIFS.2021.3098977 .
[22]
FireEye. Going ATOMIC: Clustering and associating attacker activity at scale[EB/OL]. [2022-10-15].
SONGW N, PENGG J, FUJ M, et al. Research on malicious code evolution and traceability technology[J]. Journal of Software, 2019, 30(8): 2229-2267. DOI: 10.13328/j.cnki.jos.005767(Ch ).
LÜY Q, WANGZ Y, YANGX Z, et al. A novel APT malware classification method based on feature function code[J]. Journal of Zhengzhou University (Natural Science Edition), 2023, 55(2): 10-17. DOI: 10.13705/j.issn.1671-6841.2021417(Ch ).
[27]
Zynamics. BinDiff[EB/OL]. [2022-10-15].
[28]
DUANY, LIX, WANGJ H, et al. DeepBinDiff: Learning program-wide code representations for binary diffing[C]//Proceedings of the 2020 Network and Distributed System Security Symposium. Reston: Internet Society, 2020:1-16. DOI: 10.14722/ndss.2020.24311 .
[29]
LAURENZAG, LAZZERETTIR, MAZZOTTIL. Malware triage for early identification of Advanced Persistent Threat activities[EB/OL]. 2018: 1810.07321. DOI: 10.1145/3386581 .
[30]
ALRABAEES, SHIRANIP, DEBBABIM, et al. On the feasibility of malware authorship attribution[C]//Proceedings of the 9th Foundations and Practice of Security (FPS), Cham: Springer International Publishing, 2016: 256-272. DOI:org/10.1007/978-3-319-51966-1_17 .
[31]
KUMART, SOMANIG. Origin information assisted hybrid analysis to detect APT malware[C]// Proceedings of the 17th International Conference on Information Systems Security (ICISS), Cham: Springer International Publishing, 2021: 75-93. DOI:org/10.1007/978-3-030-92571-0_5 .
[32]
SAININ, BHAT KASARAGODV, PRAKASHAK, et al. A hybrid ensemble machine learning model for detecting APT attacks based on network behavior anomaly detection[J]. Concurrency and Computation: Practice and Experience, 2023, 35(28): e7865. DOI: 10.1002/cpe.7865 .
[33]
ROSENBLUMN, ZHUX J, MILLERB P. Who Wrote This Code? Identifying the Authors of Program Binaries[M]//Proceedings of the 16th European Symposium on Research in Computer Security (ESORICS). Berlin: Springer, 2011: 172-189. DOI: 10.1007/978-3-642-23822-2_10 .
QIAOY C, YUNX C, ZHANGY Z, et al. An automatic MaIware Homology identification method based on Calling habits[J]. Acta Electronica Sinica, 2016, 44(10): 2410-2414. DOI: 10.3969/j.issn.0372-2112.2016.10.019(Ch ).
[36]
ABUHAMADM, ABUHMEDT, MOHAISENA, et al. Large-scale and language-oblivious code authorship identification[C]//Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications Security. New York: ACM, 2018: 101-114. DOI: 10.1145/3243734.3243738 .
[37]
XUAN CDO, DAOM H. A novel approach for APT attack detection based on combined deep learning model[J]. Neural Computing and Applications, 2021, 33(20): 13251-13264. DOI: 10.1007/s00521-021-05952-5 .
[38]
SHANGL K, GUOD, JIY D, et al. Discovering unknown advanced persistent threat using shared features mined by neural networks[J]. Computer Networks, 2021, 189: 107937. DOI: 10.1016/j.comnet.2021.107937 .
[39]
WEIC X, LIQ, GUOD, et al. Toward identifying APT malware through API system calls[J]. Security and Communication Networks, 2021, 2021: 8077220. DOI: 10.1155/2021/8077220 .
LIT, QIAOW, ZHANGJ W, et al. Privacy-preserving network attack provenance based on graph convolutional neural network[J]. Journal of Computer Research and Development, 2021, 58(5): 1006-1020. DOI: 10.7544/issn1000-1239.2021.20200942(Ch ).
[42]
LIZ T, CHENGX, SUNL X, et al. A hierarchical approach for advanced persistent threat detection with attention-based graph neural networks[J]. Security and Communication Networks, 2021, 2021(1): 9961342. DOI: 10.1155/2021/9961342 .
XIEL X, LIX O, YANGH Y, et al. Multi-stage detection method for APT attack based on sample feature reinforcement[J]. Journal on Communications, 2022, 43(12): 66-76. DOI:10.11959/j.issn.1000-436x.2022238(Ch ).
[45]
CHENT M, DONGC Y, LÜM Q, et al. APT-KGL: An intelligent APT detection system based on threat knowledge and heterogeneous provenance graph learning[J]. IEEE Transactions on Dependable and Secure Computing, 2022: 1-15. DOI: 10.1109/TDSC.2022.3229472 .
[46]
MILAJERDIS M, GJOMEMOR, ESHETEB, et al. HOLMES: Real-time APT detection through correlation of suspicious information flows[C]//Proceedings of the 2019 IEEE Symposium on Security and Privacy (SP). New York: IEEE Press, 2019: 1137-1152. DOI: 10.1109/SP.2019.00026 .
LIY C, LUOH, WANGX Y, et al. Construction of advanced persistent threat attack detection model based on provenance graph and attention mechanism[J]. Journal on Communications, 2024, 45(3): 117-130. DOI:10.11959/j.issn.1000-436x.2024039(Ch ).
[49]
LIS F, DONGF, XIAOX S, et al. NODLINK: An online system for fine-grained APT attack detection and investigation[C]//Proceedings of the 2024 Network and Distributed System Security Symposium. Reston: Internet Society, 2024:. DOI: 10.14722/ndss.2024.23204 .
[50]
ZHUT T, YUJ K, XIONGC L, et al. APTSHIELD: A stable, efficient and real-time APT detection system for Linux hosts[J]. IEEE Transactions on Dependable and Secure Computing, 2023, 20(6): 5247-5264. DOI: 10.1109/TDSC.2023.3243667 .