Holography: a behavior‐based profiler for malware analysis

Behavior‐based detection and signature‐based detection are two popular approaches to malware (malicious software) analysis. The security industry, such as the sector selling antivirus tools, has been using signature and heuristic‐based technologies for years. However, this approach has been proven to be inefficient in identifying unknown malware strains. On the other hand, the behavior‐based malware detection approach has a greater potential in identifying previously unknown instances of malicious software. The accuracy of this approach relies on techniques to profile and recognize accurate behavior models. Unfortunately, with the increasing complexity of malicious software and limitations of existing automatic tools, the current behavior‐based approach cannot discover many newer forms of malware either. In this paper, we implement ‘holography platform’, a behavior‐based profiler on top of a virtual machine emulator that intercepts the system processes and analyzes the CPU instructions, CPU registers, and memory. The captured information is stored in a relational database, and data mining techniques are used to extract information. We demonstrate the breadth of the ‘holography platform’ by conducting two experiments: a packed binary behavior analysis and a malvertising (malicious advertising) incident tracing. Both tasks are known to be very difficult to do efficiently using existing methods and tools. We demonstrate how the precise behavior information can be easily obtained using the ‘holography platform’ tool. With these two experiments, we show that the ‘holography platform’ can provide security researchers and automatic malware detection systems with an efficient malicious software behavior analysis solution. Copyright © 2011 John Wiley & Sons, Ltd.     

基于模型解释的PE文件对抗性恶意代码检测

深度学习已经逐渐应用于恶意代码检测并取得了不错的效果.然而,最近的研究表明:深度学习模型自身存在不安全因素,容易遭受对抗样本攻击.在不改变恶意代码原有功能的前提下,攻击者通过对恶意代码做少量修改,可以误导恶意代码检测器做出错误的决策,造成恶意代码的漏报.为防御对抗样本攻击,已有的研究工作中最常用的方法是对抗训练.然而对抗训练方法需要生成大量对抗样本加入训练集中重新训练模型,效率较低,并且防御效果受限于训练中所使用的对抗样本生成方法.为此,提出一种PE文件格式恶意代码对抗样本检测方法,针对在程序功能无关区域添加修改的一类对抗样本攻击,利用模型解释技术提取端到端恶意代码检测模型的决策依据作为特征,进而通过异常检测方法准确识别对抗样本.该方法作为恶意代码检测模型的附加模块,不需要对原有模型做修改,相较于对抗训练等其他防御方法效率更高,且具有更强的泛化能力,能够防御多种对抗样本攻击.在真实的恶意代码数据集上进行了实验,实验结果表明,该方法能够有效防御针对端到端PE文件恶意代码检测模型的对抗样本攻击.     

Metamorphic Malware Detection Using Code Metrics

ABSTRACT

Malware is becoming more and more aggressive and new techniques are emerging to allow malicious code to evade detection by antiviruses. Metamorphic malware is a particularly insidious kind of virus that changes its form at each infection. In this article, a technique for detecting metamorphic viruses is proposed that is based on identifying specific features of the assembly code, such as the instructions that change the contents of the registers, the instructions that change the control flow, and the potential code fragmentation. Such features have been derived by the analysis of a large dataset of malware. The experimentation suggests that the proposed technique produces very high precision (over 97%) in recognizing metamorphic malware, and allows also for distinguishing among different families of malware.     

一种基于主动学习的恶意代码检测方法

现有恶意代码的检测往往依赖于对足够数量样本的分析.然而新型恶意代码大量涌现,其出现之初,样本数量有限,现有方法无法迅速检测出新型恶意代码及其变种.本文在数据流依赖网络中分析进程访问行为异常度与相似度,引入了恶意代码检测估计风险,并提出一种通过最小化估计风险实现主动学习的恶意代码检测方法.该方法只需要很少比例的训练样本就可实现准确的恶意代码检测,较现有方法更适用于新型恶意代码检测.通过我们对真实的8,340个正常进程以及7,257个恶意代码进程的实验分析,相比于传统基于统计分类器的检测方法,本文方法明显地提升了恶意代码检测效果.即便在训练样本仅为总体样本数量1%的情况下,本文方法可以也可达到5.55%的错误率水平,比传统方法降低了36.5%.     

基于语义的恶意代码行为特征提取及检测方法

提出一种基于语义的恶意代码行为特征提取及检测方法,通过结合指令层的污点传播分析与行为层的语义分析,提取恶意代码的关键行为及行为间的依赖关系;然后,利用抗混淆引擎识别语义无关及语义等价行为,获取具有一定抗干扰能力的恶意代码行为特征.在此基础上,实现特征提取及检测原型系统.通过对多个恶意代码样本的分析和检测,完成了对该系统的实验验证.实验结果表明,基于上述方法提取的特征具有抗干扰能力强等特点,基于此特征的检测对恶意代码具有较好的识别能力.     

Attention-based convolutional neural network deep learning approach for robust malware classification

Recently, transforming windows files into images and its analysis using machine learning and deep learning have been considered as a state-of-the art works for malware detection and classification. This is mainly due to the fact that image-based malware detection and classification is platform independent, and the recent surge of success of deep learning model performance in image classification. Literature survey shows that convolutional neural network (CNN) deep learning methods are successfully employed for image-based windows malware classification. However, the malwares were embedded in a tiny portion in the overall image representation. Identifying and locating these affected tiny portions is important to achieve a good malware classification accuracy. In this work, a multi-headed attention based approach is integrated to a CNN to locate and identify the tiny infected regions in the overall image. A detailed investigation and analysis of the proposed method was done on a malware image dataset. The performance of the proposed multi-headed attention-based CNN approach was compared with various non-attention-CNN-based approaches on various data splits of training and testing malware image benchmark dataset. In all the data-splits, the attention-based CNN method outperformed non-attention-based CNN methods while ensuring computational efficiency. Most importantly, most of the methods show consistent performance on all the data splits of training and testing and that illuminates multi-headed attention with CNN model's generalizability to perform on the diverse datasets. With less number of trainable parameters, the proposed method has achieved an accuracy of 99% to classify the 25 malware families and performed better than the existing non-attention based methods. The proposed method can be applied on any operating system and it has the capability to detect packed malware, metamorphic malware, obfuscated malware, malware family variants, and polymorphic malware. In addition, the proposed method is malware file agnostic and avoids usual methods such as disassembly, de-compiling, de-obfuscation, or execution of the malware binary in a virtual environment in detecting malware and classifying malware into their malware family.     

面向安卓恶意软件检测的对抗攻击技术综述

随着对Android恶意软件检测精度和性能要求的提高,越来越多的Android恶意软件检测引擎使用人工智能算法.与此同时,攻击者开始尝试对Android恶意软件进行一定的修改,使得Android恶意软件可以在保留本身的功能的前提下绕过这些基于人工智能算法的检测.上述过程即是Android恶意软件检测领域的对抗攻击.本文...     

针对指令乱序变形技术的归一化研究

新出现的恶意代码大部分是在原有恶意代码基础上修改转换而来.许多变形恶意代码更能自动完成该过程,由于其特征码不固定,给传统的基于特征码检测手段带来了极大挑战.采用归一化方法,并结合使用传统检测技术是一种应对思路.本文针对指令乱序这种常用变形技术提出了相应的归一化方案.该方案先通过控制依赖分析将待测代码划分为若干基本控制块,然后依据数据依赖图调整各基本控制块中的指令顺序,使得不同变种经处理后趋向于一致的规范形式.该方案对指令乱序的两种实现手段,即跳转法和非跳转法,同时有效.最后通过模拟测试对该方案的有效性进行了验证.     

Fingerprinting Android malware families

The domination of the Android operating system in the market share of smart terminals has engendered increasing threats of malicious applications (apps). Research on Android malware detection has received considerable attention in academia and the industry. In particular, studies on malware families have been beneficial to malware detection and behavior analysis. However, identifying the characteristics of malware families and the features that can describe a particular family have been less frequently discussed in existing work. In this paper, we are motivated to explore the key features that can classify and describe the behaviors of Android malware families to enable fingerprinting the malware families with these features.We present a framework for signature-based key feature construction. In addition, we propose a frequency-based feature elimination algorithm to select the key features. Finally, we construct the fingerprints of ten malware families, including twenty key features in three categories. Results of extensive experiments using Support Vector Machine demonstrate that the malware family classification achieves an accuracy of 92% to 99%. The typical behaviors of malware families are analyzed based on the selected key features. The results demonstrate the feasibility and effectiveness of the presented algorithm and fingerprinting method.     

Malware Threat Assessment Using Fuzzy Logic Paradigm

Academicians and industrialists working on malware use static and dynamic analysis in order to understand their functionality and the menace level posed by them. Industries providing anti-malware solutions calculate malware threat level using the approaches that involve human intervention and demand the skilled analysts along with a large number of resources. With the increasing volume, velocity, and complexity of malware, assigning such a large number of resources is not possible. Thus, there is a need to develop techniques that can automatically compute the threat or damage posed by a piece of malware (to a victim machine) as soon as it appears in the wild. This assessment of damage capability level to a zero-day malware can help in providing early warnings about a specific piece of malware so that immediate attention could be paid to it in terms of allocating resources for performing a closer analysis. This paper presents an automated technique based on fuzzy modeling for computing damage potential of malicious programs, which is calculated on the basis of features obtained after performing automated analysis of malware binaries in the sandboxed environment.     

改进的基于DNN的恶意软件检测方法

当前基于深度学习的恶意软件检测技术由于模型结构及样本预处理方式不够合理等原因,大多存在泛化性较差的问题,即训练好的恶意软件检测模型对不属于训练样本集的恶意软件或新出现的恶意软件的检出效果较差。提出一种改进的基于深度神经网络（Deep Neural Network,DNN）的恶意软件检测方法,使用多个全连接层构建恶意软件检测模型,并引入定向Dropout正则化方法,在模型训练过程中对神经网络中的权重进行剪枝。在Virusshare和lynx-project样本集上的实验结果表明,与同样基于DNN的恶意软件检测模型DeepMalNet相比,改进方法对恶意PE样本集的平均预测概率提高0.048,对被加壳的正常PE样本集的平均预测概率降低0.64。改进后的方法具有更好的泛化能力,对模型训练样本集外的恶意软件的检测效果更好。     

Antivirus security: naked during updates

The security of modern computer systems heavily depends on security tools, especially on antivirus software solutions. In the anti‐malware research community, development of techniques for evading detection by antivirus software is an active research area. This has led to malware that can bypass or subvert antivirus software. The common strategies deployed include the use of obfuscated code and staged malware whose first instance (usually installer such as dropper and downloader) is not detected by the antivirus software. Increasingly, most of the modern malware are staged ones in order for them to be not detected by antivirus solutions at the early stage of intrusion. The installers then determine the method for further intrusion including antivirus bypassing techniques. Some malware target boot and/or shutdown time when antivirus software may be inactive so that they can perform their malicious activities. However, there can be another time frame where antivirus solutions may be inactive, namely, during the time of update. All antivirus software share a unique characteristic that they must be updated at a very high frequency to provide up‐to‐date protection of their system. In this paper, we suggest a novel attack vector that targets antivirus updates and show practical examples of how a system and antivirus software itself can be compromised during the update of antivirus software. Local privilege escalation using this vulnerability is also described. We have investigated this design vulnerability with several of the major antivirus software products such as Avira, AVG, McAfee, Microsoft, and Symantec and found that they are vulnerable to this new attack vector. The paper also discusses possible solutions that can be used to mitigate the attack in the existing versions of the antivirus software as well as in the future ones. Copyright © 2013 John Wiley & Sons, Ltd.     

基于数据特征的内核恶意软件检测

内核恶意软件对操作系统的安全造成了严重威胁.现有的内核恶意软件检测方法主要从代码角度出发,无法检测代码复用、代码混淆攻击,且少量检测数据篡改攻击的方法因不变量特征有限导致检测能力受限.针对这些问题,提出了一种基于数据特征的内核恶意软件检测方法,通过分析内核运行过程中内核数据对象的访问过程,构建了内核数据对象访问模型;然后,基于该模型讨论了构建数据特征的过程,采用动态监控和静态分析相结合的方法识别内核数据对象,利用EPT监控内存访问操作构建数据特征;最后讨论了基于数据特征的内核恶意软件检测算法.在此基础上,实现了内核恶意软件检测原型系统MDS-DCB,并通过实验评测MDS-DCB的有效性和性能.实验结果表明：MDS-DCB能够有效检测内核恶意软件,且性能开销在可接受的范围内.     

Windows平台恶意软件智能检测综述

近年来,恶意软件给信息技术的发展带来了很多负面的影响.为了解决这一问题,如何有效检测恶意软件则一直备受关注.随着人工智能的迅速发展,机器学习与深度学习技术逐渐被引入到恶意软件的检测中,这类技术称之为恶意软件智能检测技术.相比于传统的检测方法,由于人工智能技术的应用,智能检测技术不需要人工制定检测规则.此外,具有更强的泛...     

恶意代码的函数调用图相似性分析

恶意代码的相似性分析是当前恶意代码自动分析的重要部分。提出了一种基于函数调用图的恶意代码相似性分析方法,通过函数调用图的相似性距离ＳＤＭＦＧ来度量两个恶意代码函数调用图的相似性,进而分析得到恶意代码的相似性,提高了恶意代码相似性分析的准确性,为恶意代码的同源及演化特性分析研究与恶意代码的检测和防范提供了有力支持。     

基于环境敏感分析的恶意代码脱壳方法

加壳技术是软件的常用保护手段,但也常被恶意代码用于躲避杀毒软件的检测.通用脱壳工具根据加壳恶意代码运行时的行为特征或统计特征进行脱壳,需要建立监控环境,因此易受环境敏感技术的干扰.文中提出了一种基于环境敏感分析的恶意代码脱壳方法,利用动静结合的分析技术检测并清除恶意代码的环境敏感性.首先,利用中间语言对恶意代码的执行轨迹进行形式化表示;然后,分析执行轨迹中环境敏感数据的来源和传播过程,提取脱壳行为的环境约束;最后,求解环境约束条件,根据求解结果对恶意代码进行二进制代码插装,清除其环境敏感性.基于此方法,作者实现了一个通用的恶意代码脱壳工具:MalUnpack,并对321个最新的恶意代码样本进行了对比实验.实验结果表明MalUnpack能有效对抗恶意代码的环境敏感技术,其脱壳率达到了89.1%,显著高于现有基于动态监控的通用脱壳工具的35.5%和基于特征的定向脱壳工具的28.0%.     

Attention-CNN在恶意代码检测中的应用研究

恶意代码攻击已经成为互联网最重要的威胁之一,并且现存的恶意代码数据庞大,特征多样.为了更好地提取恶意代码特征以及掌握恶意代码的行为,提出了基于注意力机制的Attention-CNN恶意代码检测模型.首先结合卷积神经网络(CNN)和注意力机制,构建了Attention-CNN恶意代码检测模型;然后将恶意代码转化为灰度图像...     

一种恶意软件行为分析系统的设计与实现

基于虚拟化技术的恶意软件行为分析是近年来出现的分析恶意软件的方法。该方法利用虚拟化平台良好的隔离性和控制力对恶意软件运行时的行为进行分析,但存在两方面的不足：一方面,现有虚拟机监视器（Virtual Machine Monitor,VMM）的设计初衷是提高虚拟化系统的通用性和高效性,并没有充分考虑虚拟化系统的透明性,导致现有的VMM很容易被恶意软件的环境感知测试所发现。为此,提出一种基于硬件辅助虚拟化技术的恶意软件行为分析系统——THVA。THVA是一个利用了安全虚拟机（SVM）、二级页表（NPT）和虚拟机自省等多种虚拟化技术完成的、专门针对恶意软件行为分析的微型VMM。实验结果表明,THVA在行为监控和反恶意软件检测方面表现良好。     

一种基于混合特征的移动终端恶意软件检测方法

随着移动终端恶意软件的种类和数量不断增大,本文针对Android系统恶意软件单特征检测不全面、误报率高等技术问题,提出一种基于动静混合特征的移动终端恶意软件检测方法,以提高检测的覆盖率、准确率和效率。该方法首先采用基于改进的CHI方法和凝聚层次聚类算法优化的K-Means方法构建高危权限和敏感API库,然后分别从静态分析和动态分析两个方面提取移动终端系统混合特征。在静态分析中,首先反编译APK文件,分析得到权限申请特征和敏感API调用特征;在动态分析中,通过实时监控APP运行期间的动态行为特征,分别提取其在运行过程中的敏感API调用频次特征和系统状态等特征信息;接着分别使用离差标准化、TF-IDF权重分析法和优序图法对混合特征进行归一化和特征权重赋值处理。最后,通过构建测评指标对本文所提基于混合特征恶意软件检测方法进行对比测试验证和评价分析。实验结果表明:本方法针对Android系统恶意软件的检测具有好的准确率和效率,可有效提高移动终端恶意软件检测的精确度。     

基于图模式与内存足迹的Android恶意应用与行为检测

现有的各个Android应用商店大多检查已知的静态恶意应用,难以检测新颖、动态加载的恶意应用与行为,对此提出一种基于图结构与内存足迹分析的恶意应用检测系统。首先,采集应用的内存信息,分析应用的足迹与序列号,检测动态打包的恶意代码与新颖的恶意应用;然后,提取应用所请求的共生权限,将权限建模为图结构,并使用图的度量指标分析图的分类模式与中心权限,根据中心权限值选择可表示各类的最优图指标;最终,计算应用的隐私分数与风险阈值,基于该阈值检测各种恶意软件或恶意行为。仿真实验的结果表明,本算法对不同类型的恶意应用均具有较好的效果,对于未知的恶意应用也具有较好的检测率。