Enter Sandbox: Android Sandbox Comparison

Enter Sandbox: Android Sandbox Comparison
Sebastian Neuner, Victor van der Veeny, Martina Lindorferz, Markus Huber,
Georg Merzdovnik, Martin Mulazzani and Edgar Weippl
SBA Research
Email: fsneuner,mhuber,gmerzdovnik,mmulazzani,eweipplg@sba-research.org
yThe Network Institute, VU University Amsterdam; ITQ
Email: v.vander.veen@vu.nl
zSecure Systems Lab, Vienna University of Technology
Email: mlindorfer@iseclab.org
Abstract—Expecting the shipment of 1 billion Android devices
in 2017, cyber criminals have naturally extended their vicious
activities towards Google’s mobile operating system. With an
estimated number of 700 new Android applications released every
day, keeping control over malware is an increasingly challenging
task. In recent years, a vast number of static and dynamic code
analysis platforms for analyzing Android applications and making
decision regarding their maliciousness have been introduced in
academia and in the commercial world. These platforms differ
heavily in terms of feature support and application properties
being analyzed. In this paper, we give an overview of the state-of-the-
art dynamic code analysis platforms for Android and evaluate
their effectiveness with samples from known malware corpora
as well as known Android bugs like Master Key. Our results
indicate a low level of diversity in analysis platforms resulting
from code reuse that leaves the evaluated systems vulnerable to
evasion. Furthermore the Master Key bugs could be exploited by
malware to hide malicious behavior from the sandboxes.
Keywords—Android, malware, dynamic analysis, sandbox eva-sion,
sandbox fingerprinting
I. INTRODUCTION
With an estimated market share of 70% to 80%, Android
has become the most popular operating system for smartphones
and tablets [1, 2]. Expecting a shipment of 1 billion Android
devices in 2017 and with over 50 billion total app downloads
since the first Android phone was released in 2008, cyber
criminals naturally expanded their vicious activities towards
Google’s mobile platform. In the summer of 2012, the so-phisticated
Eurograbber attack showed that mobile malware
may be a very lucrative business by stealing an estimated
EUR 36,000,000 from bank customers in Italy, Germany, Spain
and the Netherlands [3]. Also in 2012 the mobile malware
SMSZombie [4] infected over 500,000 devices. In addition
to obtaining administrator privileges, SMSZombie also gained
control over the mobile SMS payment system of China Mobile,
enabling the malware authors to secretly authorize payments.
An additional incentive for mobile malware authors to
target Android instead of other mobile platforms is Android’s
open design that allows users to install applications from a
variety of sources. However, the diversity of third-party app
stores and volume of apps published poses a considerable
challenge to security researchers and app store administrators
when trying to identify malicious applications. With over 1
million apps available for download via the official Google
Play Store [5] alone, and possibly another million spread
among third-party app stores, it is possible to estimate that over
20,000 new applications are being released every month [6].
This requires scalable solutions for quickly analyzing new
apps in order to isolate malicious and other possibly unwanted
apps. Google reacted to the growing interest of miscreants in
Android by introducing Bouncer in February 2012, a dynamic
analysis sandbox that automatically checks apps submitted to
the Google Play Store for malware [7]. However, research has
shown that Bouncer’s detection rate is still fairly low and that
it can be easily bypassed [8, 9].
More than a year before Google’s deployment of Bouncer,
Bl¨asing et al. [10] were the first to present a dynamic analysis
platform for Android applications called AASandbox (Android
Application Sandbox). Since then, an ever increasing number
of dynamic sandboxes for analyzing potentially malicious
Android applications have been introduced in academia and in
the commercial world. Similar to dynamic malware analysis
platforms for Windows binaries such as CWSandbox [11] or
Anubis [12], dynamic sandboxes for Android run targeted
applications in a controlled environment to obtain a behavioral
footprint. Results are normally presented to the user in the form
of a report, possibly including a classification that indicates
whether the app is benign or malicious. Most of these systems
use some form of hybrid analysis, i.e., leveraging additional
static analysis during a preprocessing phase to enhance the
dynamic analysis results.
Egele et al. [13] systemized the existing knowledge on
dynamic analysis of traditional malware. However, no such
survey on dynamic malware analysis in a mobile context exists.
In this paper, we systematically analyze the state-of-the-art
dynamic analysis approaches for the Android platform and
compare existing approaches in terms of provided features and
analysis effectiveness.
In particular, the contributions of this paper are as follows:
We survey current state-of-the-art Android malware
detection techniques.
We discuss methods to detect and fingerprint dynamic
analysis sandboxes.
We compare 16 dynamic analysis platforms for An-droid
regarding their features, level of introspection,
functionality and interdependencies.

We evaluate the effectiveness of ten of these dynamic
sandboxes using a selected set of malware samples
from publicly available malware corpora. Furthermore,
we analyze the susceptibility of these dynamic sand-boxes
to the so-called Master Key vulnerabilities.
The remainder of this paper is organized as follows:
Section II gives an overview of current Android malware be-havior,
as well as distribution techniques. Section III discusses
Android analysis techniques and numerous existing malware
analysis frameworks. We then discuss our evaluation criteria
and sandbox interdependencies in Section IV. Finally, we
compare the sandboxes and evaluate their effectiveness and
limitations in Section V, before we conclude in Section VI.
II. ANDROID MALWARE
In order to gain a better understanding of the requirements
for Android malware detection techniques to successfully
analyze mobile malware samples, we give a brief overview on
the current mobile malware threat landscape. In the following
we discuss the motivation of mobile malware authors, different
distribution methods as well as available malware data sets.
A. Motivation
In August 2010 the first malicious Android application
named AndroidOS.FakePlayer was discovered in the wild [14].
FakePlayer monetized infected devices by secretly sending
text messages to premium numbers. Since then, both the
sophistication as well as the amount of observed malware
samples increased steadily. Recent reports focusing on mobile
malware trends estimate that the number of malicious Android
apps now ranges from 120,000 to 718,000 [15, 16].
The main motivation for mobile malware authors is finan-cial
gain. One way to monetize infected devices, leveraged by
FakePlayer and countless other malware families since then,
is by sending text messages to premium numbers, registered
to malware authors. In addition to these so called toll fraud
schemes, malware authors leverage their apps to spy on
users and collect personal information. Mobile spyware has
capabilities to forward private data to a remote server under
the control of malware authors. In a more complex form, the
malware could also receive commands from the server to start
specific activities and become part of a botnet. Furthermore,
mobile versions of the banking Trojan ZeuS (ZeuS-in-the-
Mobile, or ZitMo) are a way to circumvent the two-factor
authentication of online banking systems by stealing mobile
TAN codes [17]. Broadcast receivers are an Android-specific
feature of particular interest to malware authors as they can be
used to launch a background service as soon as the device
is started and secretly intercept and forward incoming text
messages to a remote server. This capability was used in the
aforementioned Eurograbber attack [3] in order to authorize
financial transactions.
B. Distribution
To lure victims into installing malicious apps, a common
strategy employed by malware authors is to repackage popular
applications with malicious payloads. The target applications
often include paid applications, which are then offered for
“free”. Attackers commonly use third-party marketplaces to
distribute their repackaged applications, as these marketplaces
fail to verify submitted apps. Juniper Networks, for instance,
found that malicious applications often originate from third-party
marketplaces, with China and Russia being the world’s
leading suppliers [18]. Zhou et al. [19] analyzed repackaged
apps in six alternative Android marketplaces and found that
in addition to repackaging apps with malicious payloads,
repackaged apps furthermore modify embedded advertising
libraries to steal the ad-revenue of application developers.
Attackers can also leverage drive-by downloads adapted to
the mobile context. Methods to trick users into installing ma-licious
apps include in-app advertising [20], specially crafted
websites or QR codes [21]. Drive-by downloads might however
also exploit platform-level vulnerabilities to install malware
in a stealth fashion. Due to Android’s loose management to
device software, Android versions have become fragmented,
with only 1.8% of all devices running the latest Android
version 4.4 (codename KitKat) [22], as of February 2014.
This fragmentation makes new security features, as well as
bugfixes for core components preventing against arbitrary code
execution exploits, only available to a small group of users.
Android versions prior to 2.3.7 are especially vulnerable to
root exploits (examples include RageAgainstTheCage [23],
Exploid [24] and zergRush [25]). While these exploits were
originally developed to overcome limitations that carriers and
hardware manufactures put on some devices, they have also
been used by malware authors to obtain a higher privilege
level without a user’s consent. This approach allows malware
to request only a few permissions during app installation, but
still gaining root access to the entire system once the app is
executed.
C. Malware Data Sets
Access to known Android malware samples is mainly
provided via the Android Malware Genome Project [26],
Contagio Mobile [27] and VirusShare [28]. The Android Mal-ware
Genome Project contains over 1,200 Android malware
samples from 49 families, collected from August 2010 to
October 2011. Contagio Mobile offers an upload dropbox to
share mobile malware samples among security researchers and
currently hosts 164 archives, where some archives contain
more than 16,000 samples. VirusShare also hosts a reposi-tory
of malware samples with over 11,000 Android samples
available to researchers. Furthermore, the multi-engine anti-virus
scanning services VirusTotal [29] and AndroTotal [30]
provide researchers with access to submitted samples.
III. ANDROID MALWARE ANALYSIS FRAMEWORKS
Ever since the first Android phones were released in
2008, researchers have proposed dozens of frameworks for a
variety of purposes. In this section, we outline our efforts in
systematizing these proposals by enumerating and analyzing
this existing knowledge.
Analysis frameworks may use a number of techniques to
produce a report about an app’s functionality or perform a
classification whether an app is benign or malicious. Static
analysis techniques extract features from the Android appli-cation
package (APK) and the Dalvik bytecode. Dynamic

analysis techniques monitor an app’s behavior during runtime
in a controlled environment. Results from static analysis can
also enhance dynamic analysis, e.g. to efficiently stimulate a
targeted application and trigger additional behavior, resulting
in hybrid analysis approaches.
In the following paragraphs we briefly explain different
analysis methods for both static and dynamic analysis and
present available tools and frameworks for both approaches.
Our evaluation mainly focuses on dynamic analysis frame-works,
however, we also present various static analysis tools
that might assist dynamic analysis and therefore be inte-grated
in the presented dynamic analysis frameworks. For
each dynamic analysis framework, we distinguish a number
of characteristics and provide a brief summary. Based on their
main purpose and approach, we classify the different research
efforts into distinct categories.
A. Static Analysis Tools
Static analysis tools may fall in one of the following
categories:
Extraction of meta information: Tools that extract
information from an application’s manifest and pro-vide
information about requested permissions, activi-ties,
services and registered broadcast receivers. Meta
information is often used during later dynamic analy-sis
in order to trigger an application’s functionality.
Weaving: Tools that rewrite bytecode of existing ap-plications
using a bytecode weaving technique. Using
this technique, analysis frameworks can, for instance,
insert tracing functionality into an existing application.
Decompiler: Tools that implement a Dalvik bytecode
decompiler or disassembler.
One of the most popular comprehensive static analysis tool
for Android applications is Androguard [31]. It can disas-semble
and decompile Dalvik bytecode back to Java source
code. Given two APK files, it can also compute a similarity
value to detect repackaged apps or known malware. It also
has modules that can parse and retrieve information from the
app’s manifest. Due to its flexibility, it is used by some other
(dynamic) analysis frameworks that need to perform some
form of static analysis. APKinspector1 is a static analysis plat-form
for Android application analysts and reverse engineers to
visualize compiled Android packages and their corresponding
DEX code. Dexter2 is a web application designed for static
analysis of Android applications. Its features are comparable
with those of Androguard and APKinspector, however, it has
some additional collaboration functionality to ease knowledge
sharing among multiple researchers.
APKtool3 is a tool for reverse engineering Android ap-plications.
It can decode resources to nearly original form
and rebuild them into a new Android package after they
have been modified. APKtool can be used to add additional
features or extra support to existing applications without con-tacting
the original author and thus assist bytecode weaving
1https://github.com/honeynet/apkinspector/
2http://dexter.dexlabs.org/
3https://code.google.com/p/android-apktool/
approaches. Joe Sandbox Mobile uses static instrumentation,
which is equivalent to bytecode weaving in our classification.
Radare24 is an open source reverse engineering framework
which provides a set of tools to disassemble, debug, analyze,
and manipulate binary Android files. Other disassembler tools
for DEX files include Dedexer5 and smali/baksmali6. Both read
DEX files and convert them into an ‘assembly-like’ format
which is largely influenced by the Jasmin syntax7.
Other tools aim at enabling static analysis on Android
applications by retargeting them to traditional .class files,
which then can be processed by existing Java tools. One
example is ded [32, 33], which later evolved into Dare8.
Similarly, dex2jar9 can convert an APK file directly to a .jar
file and vice versa. JEB10 is a commercial flexible interactive
Android decompiler. It claims to be able to directly decompile
Dalvik bytecode to Java source code, as well as disassemble
an APK’s contents so that users can view the decompressed
manifest, resources, certificates, etc.
B. Dynamic/Hybrid Analysis Frameworks
Dynamic analysis frameworks monitor the behavior of
unknown applications at runtime by executing the targeted ap-plication
in a controlled environment to generate a behavioral
footprint. Dynamic analysis can monitor an app’s behavior
utilizing one (or more) of the following techniques:
Taint tracking: Taint tracking tools are often used in
dynamic analysis frameworks to implement system-wide
dynamic taint propagation in order to detect
potential misuse of users’ private information.
Virtual machine introspection (VMI): VMI-based
frameworks [34] intercept events that occur within
the emulated environment. Dalvik VMI based systems
monitor the execution of Android APIs through modi-fications
in the Dalvik VM. Qemu VMI based systems
are implemented on the emulator level to enable the
analysis of native code. However, emulators are prone
to emulator evasion [35].
System call monitoring: Frameworks can collect
an overview of executed system calls, by using, for
instance, VMI, strace or a kernel module. This
enables (partial) tracing of native code.
Method tracing: Frameworks can trace Java method
invocations of an app in the Dalvik VM.
The first dynamic Android analysis framework developed
was AASandbox [10], presented in October 2010. It imple-ments
a system call monitoring approach using a loadable
kernel module. Furthermore, it uses the resulting system call
footprint to discover possibly malicious applications.
A popular taint tracking framework is TaintDroid [36].
TaintDroid is implemented upon the Dalvik VM and monitors
4http://radare.org/y/
5http://dedexer.sourceforge.net
6https://code.google.com/p/smali/
7Jasmin is an assembler for the Java VM: http://jasmin.sourceforge.net
8http://siis.cse.psu.edu/dare/
9http://code.google.com/p/dex2jar/
10http://www.android-decompiler.com/

applications for the leakage of sensitive information. However,
ScrubDroid [37] presented a number of attacks to circumvent
dynamic taint analysis. VetDroid [38] is a dynamic framework
that measures actual permission use behavior by dynamically
building a permission use graph. Their data tainting method is
built upon TaintDroid, but improves it by identifying implicit
and explicit permission use points.
DroidBox [39] uses TaintDroid to detect privacy leaks,
but also comes with a Dalvik VM patch to monitor the
Android API and report file system and network activity, the
use of cryptographic operations and cell phone usage such
as sending SMS and making phone calls. However, it can
be easily bypassed if applications include their own libraries,
in particular by using it for cryptographic operations. In the
latest version, DroidBox utilizes bytecode weaving to insert
monitoring code into the app under analysis11. ANANAS [40]
is a recently published framework focusing on a modularized
architecture with analysis modules for logging e.g. file system
activity, network activity or system calls. It also employs API
monitoring similar to the most recent version of DroidBox and
uses a loadable kernel module for system call monitoring.
DroidScope [41] uses VMI to reconstruct Dalvik and native
code instruction traces. The authors also implemented their
own data tainting method, named TaintTracker. One of the
huge benefits is that it does not require any changes to the
Android sources, however JIT has to be selectively disabled
as it blurs the Dalvik instruction boundaries. Andrubis [42]
leverages various techniques, is VMI-based and monitors
events in the Dalvik VM as well as native code for system-calls
through QEMU VMI. It further employs TaintDroid for
data tainting. TraceDroid [43] has the benefit of generating
a complete method trace output and was also integrated into
Andrubis. It further captures network traffic and refrains from
analyzing native code, as ltrace and strace are deemed
sufficient for this purpose. This is also the approach used by
Mobile Sandbox [44], which runs the app in the emulator and
uses ltrace to track native code. Finally, CopperDroid [45]
uses a VMI-based approach as well, but compared to the other
frameworks it is also capable of analyzing IPC and RPC-based
communication between applications.
Andrubis, TraceDroid, Mobile Sandbox and CopperDroid
all perform input stimulation, most notably the monkey exer-ciser
and the emulation of common events like GPS lock, SMS
received or boot-completed. To further enhance code coverage,
several frameworks for automatically extracting and exercising
UI based triggers have been proposed. The SmartDroid frame-work
[46] statically extracts the function call graph and activity
call graph, and then dynamically traverses these graphs to find
elements that trigger sensitive behavior. In contrast to other
frameworks, however, it does not focus on possibly malicious
activities like sending SMS or accessing files. AppsPlay-ground
[47] is a framework that automates analysis of Android
applications and monitors taint propagation (using TaintDroid),
specific API calls and system calls. Its main contribution is
a heuristic-based intelligent black-box execution approach to
explore the app’s GUI. The goal of AppIntent [48] is to distin-guish
user-intended data transmission from unintended ones,
and is as such related to SmartDroid. It uses an event-space
11http://code.google.com/p/droidbox/wiki/APIMonitor
constraint guided symbolic execution technique to construct
event sequences, which effectively reduces the event search
space in symbolic execution for Android apps.
A number of additional dynamic analysis platforms have
been implemented and made available to the public via web
applications: SandDroid12, VisualThreat13, ForeSafe14 and the
Joe Sandbox Mobile APK Analyzer15. These frameworks,
however, come often with very little public documentation on
how they operate, which makes it hard to make statements
on any new approaches used by these implementations. It is
likely that most of these platforms use (modified versions of)
existing tools like DroidBox, TaintDroid and Androguard to
complement their dynamic analysis engine. This is confirmed
for example on SandDroid’s webpage, which states that it is
powered by both DroidBox and Androguard. While we could
not find a public reference which Android version is used by
Joe Sandbox Mobile, it was possible to create APK files with
varying API level requirements and fingerprint the Android
version to API level 15. This either is Android version 4.0.3 or
4.0.4, which is an upper bound of supported Android versions.
Unfortunately, this fingerprinting technique was unsuccessful
with ForeSafe and VisualThreat.
C. Miscellaneous Frameworks
For the sake of completeness it is also worth mentioning
various other frameworks that offer online services for the
analysis of Android apps based on their static features or meta
information: Badger16 provides a web service for extracting
required permissions as well as included advertising applica-tions
from uploaded APK files. Undroid17 performs slightly
more comprehensive static analysis and extracts requested per-missions
as well as API calls that use these permissions. The
Android Observatory [49] also extracts required permissions
from the manifest and further tries to match the app’s certificate
against those of known malware.
Finally, VirusTotal [29] and AndroTotal [30] are two multi-engine
anti-virus scanning services. VirusTotal uses static
AV-based scanner engines to assess the maliciousness of a
submitted APK file. AndroTotal uses dynamic sandboxing
to test samples against mobile malware detectors, similar to
VirusTotal. However, it uses the apps themselves on physical
devices as well as in the emulator instead of command line
versions with the same underlying signature database. It also
provides links to analysis results from other frameworks:
VirusTotal, CopperDroid, ForSafe, SandDroid and Andrubis.
IV. EVALUATION CRITERIA
We present an overview of all dynamic analysis frame-works
discussed so far in Table I and list, whether the source
code is publicly available (src) or the service is available
through a web interface (www).
12http://sanddroid.xjtu.edu.cn
13http://www.visualthreat.com
14http://www.foresafe.com
15http://www.apk-analyzer.net
16http://davidson-www.cs.wisc.edu/baa
17http://www.av-comparatives.org/avc-analyzer/

TaintDroid Monkey Androguard
TraceDroid
ANANAS Mobile Sandbox SandDroid Andrubis CopperDroid
AppsPlayground
DroidBox vetDroid
AASandbox
ForeSafe
AppIntent
DroidScope
Joe Sandbox
Mobile
SmartDroid
VisualThreat
Fig. 1. Interdependency graph illustrating sandboxes relying on other tools or services.
TABLE I. OVERVIEW OF ANDROID ANALYSIS FRAMEWORKS AND THEIR
AVAILABILITY - EITHER AS SOURCE CODE (SRC) OR THROUGH A PUBLIC
WEB INTERFACE (WWW).
Framework src www Framework src www
AASandbox [10] ForeSafe
AppIntent [48] Joe Sandbox Mobile
ANANAS [40] Mobile Sandbox [44]
AndroTotal [30] SandDroid
Andrubis [42] SmartDroid [46]
AppsPlayground [47] TaintDroid [36]
CopperDroid [45] TraceDroid [43]
DroidBox [39] vetDroid [38]
DroidScope [41] VisualThreat
A. Features of Interest
In order to compare the discussed dynamic analysis frame-works
we manually examined them and compiled a feature
list for further evaluation. First of all, we are interested in
implementation details such as the Android version that is
supported by the sandbox. Android is under active devel-opment,
and at the time of writing Android 4.4 (codename
KitKat) is the most recent version. We are also interested in
the inspection level and which methods the sandboxes use
to capture the dynamic behavior of an app. As discussed in
Section III-B, multiple methods are suitable: Modifying stock
Android for additional logging capabilities is one convenient
way to capture app behavior, in particular the kernel or the
APIs. ANANAS for example uses a loadable kernel module
for its analysis. The downside of this approach is that the kernel
patches cannot be easily ported to other versions of Android,
as the internals can change between updates. Secondly, as
Android programs are executed within the Dalvik VM, some
approaches modify the Dalvik VM to capture the dynamic
behavior of an app. TaintDroid and DroidBox use this method
for example. Downside of this approach is that only Java calls
are traceable and native code is not captured. Lastly, it is
possible to modify QEMU for enhanced logging mechanisms
and to use true virtual machine introspection. This is a holistic
method, as it works on a lower level compared to the other
methods. However, arm-to-x86 emulation is known to be slow,
and as such it has a high overhead. Some sandboxes even use
combinations of various methods e.g., Andrubis uses virtual
machine introspection through QEMU in combination with a
modified Dalvik VM.
Additionally, some sandboxes use taint tracking to detect if
the users private information is leaked. Furthermore, sandboxes
usually perform some form of static analysis such as parsing
the requested permissions from the manifest. GUI interactions
are usually employed to simulate user activity and enhance
the coverage of dynamic analysis. We can further distinguish
the sandboxes by the features they analyze. Network activity
as well as file activity is of particular interest in malware
analysis, as malware usually reads and writes local files and
relies on obtaining commands from a command and control
server in case of a botnet [50]. As discussed in Section II-A,
malware is often used to generate revenue for the malware
creator by sending premium SMS or calling premium phone
numbers causing financial damage to the victim. As such, we
are interested if the sandbox is able to capture phone activity.
Furthermore, we are interested if the analysis frameworks are
able to analyze native code.
B. Framework Interdependency
Many sandboxes build on previous work and incorporate
existing tools in their service. In particular a combination of the
open source solutions TaintDroid and DroidBox for dynamic
analysis and Androguard for static analysis are heavily used.
As we will show in the evaluation in Section V, this can be a
problem if the malware is capable of evading these underlying
analysis tools. Furthermore, many sandboxes rely on the Appli-cation
Exerciser Monkey18 (or Monkey) to simulate user input,
even though the intended use case of Monkey was to stress test
applications with pseudorandom user events. Figure 1 shows
the full interdependency graph depicting which sandboxes rely
on which other tools or services.
C. Sandbox Fingerprinting
Dynamic sandboxes are often rather easy to fingerprint,
which is a problem that has been extensively studied for
PC malware [51, 52, 53]. First of all, system emulators like
18http://developer.android.com/tools/help/monkey.html

QEMU [35, 54] and virtual machines [55] are detectable.
With the numerous sensors and interfaces of a smartphone,
a real device is even more challenging to simulate than PC
hardware. GPS, screen size, and the motion sensor are just
a few examples of what has to be considered when building
sandboxes for mobile devices. Malware can also simply wait
for a certain amount of time, as dynamic sandboxes cannot
execute an app indefinitely [56] due to limited resources. Other
methods to detect if the app is running within a sandbox
include checking the application’s signature. This makes it
possible to detect platforms like the recent version of DroidBox
that modify the app’s bytecode.
Fingerprinting methods can also rely on implementation
details of specific sandboxes. For example, AASandbox uses
500 monkey events to simulate user input, with 1 second
between events [10]. While this is of course just a configuration
parameter, malware could easily detect AASandbox if these
parameters are not modified. TraceDroid on the other hand
initializes the Monkey with the same seed value for the pseudo-random
number generator (currently set to 1337) [43]. This is
intended to ensure that re-runs of the exerciser generate the
same sequence of events, but this is however also detectable
by malware. SmartDroid [46] waits for 10 seconds before
starting UI interactions during dynamic analysis. There is yet
no ground truth on the distribution and statistical likelihood of
user events.
V. EVALUATION RESULTS
We evaluated a total of 10 dynamic sandboxes based on
the feature set outlined above, skipping sandboxes that are not
available to us in any form. Our analysis is based on the aca-demic
publications as well as available documentation. Table II
lists the extracted features for all sandboxes. We furthermore
tested the sandboxes with selected malware samples, which we
discuss in the following paragraphs.
A. Malware Samples
For our evaluation we used real-world malware samples
from four different families or categories. For each of those
families we used two different samples. All of the samples
have been publicly analyzed and described, either as part of
the Android Malware Genome Project [26] or in blogposts or
technical reports from antivirus vendors. All four families were
chosen due to either feature coverage (regarding [26] coverage
in terms of financial charges and personal information stealing)
or due to interesting behaviour like sophisticated privilege
escalation techniques or evasion of specific sandboxes.
a) Obad: In 2013, Kaspersky Lab [57] described Obad
as one of the most sophisticated mobile malware to date. The
user has to install the malware manually with 24 different
permissions, which grant nearly total access to the device.
Obad is able to send SMS, send data to an URL, download and
install files, and transfer files via bluetooth. Additionally, the
application tries to connect to a command and control server
and is able to join a botnet. Malware of the Obad family tries
to evade detection from several sandboxes with anti-emulation
and anti-decompilation techniques, like checking for the An-droid.
os.build.MODEL value to quit the applications execution
if the default value of the emulator is present [58, 59].
b) Geinimi: The second Android malware family we
selected for our analysis is Geinimi, which is part of the
Android Malware Genome Project. We selected Geinimi due
to its various malicious payloads: remote control over the
Internet, starting phone calls, sending SMS as well as leaking
sensitive data stored on the phone.
c) DroidKungFu: DroidKungFu [60] is part of the
Android Malware Genome Project as well, and is a malware
family that can perform various forms of privilege escalation.
It collects various phone related data such as the IMEI number,
the phone model etc. and sends it to a server. This malware
family is of particular interest since it is able to execute
various exploits, for example privilege escalation using the
RageAgainstTheCage exploit [23]. This exploit decrypts itself
during runtime to exploit the adb resource exhaustion bug for
root access. After this procedure the application is able gain
root privileges and hide itself, drop other malware, etc.
d) Basebridge/Nyleaker: The fourth malware category
we used for our evaluation are the malware samples Basebridge
and Nyleaker. Both samples are able to successfully evade
analysis by abusing the Androguard tool (which is used by
many sandboxes) by presenting itself as an invalid application
by either using a corrupt APK or an invalid manifest file within
the APK. Their malicious behavior includes sending SMS to
premium services, executing local privilege escalation exploits
and leaking personal information.
e) Master Key: The last evaluated category is the
Master Key set that exploits several Android vulnerabilities
related to the handling of ZIP files19.
Bug 821932120: The ZIP format allows multiple oc-currences
of the same filename within one archive.
This can lead to a serious vulnerability, if the imple-mentation
of the unpacking of the archive differs in
parts of the systems: The Android signature verifier
checks the first file for integrity and compares it
with the META-INF directory. The Android installer
however uses the second file for installation. Symantec
reported that this bug was already exploited by mal-ware
in the wild [61]. Specifically malware samples
added an additional malicious classes.dex as well as a
second manifest file to a benign app without breaking
its signature.
Bug 969586021: This bug is a simple signed-unsigned
integer mismatch between different parts of the An-droid
code. The ZIP file header contains the fields for
filename length and extra field length. The signature
verification code treats the contents of these fields
as signed 16-bit integers, which converts the input
of ”0xFFFD” to ”-3”. Thus, the Android verifier
jumps three bytes backwards, instead of forward and
therefore skips the extra field by reading the part
”dex” of ”classes.dex”. The application loading code
however treats the fields as unsigned 16-bit integers,
19APK files are ZIP compressed files, based on the JAR format.
20http://nakedsecurity.sophos.com/2013/07/10/anatomy-of-a-security-hole-googles-
android-master-key-debacle-explained
21http://nakedsecurity.sophos.com/2013/07/17/anatomy-of-another-android-
hole-chinese-researchers-claim-new-code-verification-bypass/

TABLE II. COMPARISON OF ANDROID MALWARE ANALYSIS SANDBOXES.
Implementation Details Analysis Type Analyzed Features
Framework Android Version Inspection Level Static Tainting GUI Interactions File Network Phone Native Code
AASandbox — Kernel
AppIntent 2.3 Kernel
ANANAS 2.3-4.2 Kernel
Andrubis 2.3.4 QEMU Dalvik
AppsPlayground — Kernel
CopperDroid 2.2.3 QEMU
DroidBox 2.3-4.1 Kernel
DroidScope 2.3 Kernel Dalvik
ForeSafe ? ?
Joe Sandbox Mobile 4.0.3 / 4.0.4 Static Instrumentation
Mobile Sandbox 2.3.4 Dalvik
SandDroid ? ? ? ? ?
SmartDroid 2.3.3 Kernel
TraceDroid 2.3.4 Dalvik
vetDroid 2.3 Kernel Dalvik
VisualThreat ? ?
which could match the start of a second uncompressed
dex file supplied in the file data section.
Bug 995069722: This last version of the Master Key
vulnerability is caused by the redundant storage of
the ”filename length” field in the ZIP header. This
field indicates how many bytes between the filename
and the actual file data exist and also how many
bytes in the central directory exist, to reach the next
directory entry in the archive. It is possible to provide
a real filename length for the verifier, that verifies
the trustworthy original file data and fake a filename
length in the local header for the loader, that later
executes the malware code.
Additionally, a certain bug in Python23 can be used to
create APK files which are not processed correctly by certain
sandboxes. This Python bug triggers an exception when the
length field in the ZIP header is zero. While this is a Python
issue and not related directly to Android, many tools and
sandboxes which are based on Python are affected and malware
authors could craft APK files that trigger this bug to evade
analysis.
B. Analysis Results
We submitted the 12 malware samples (8 from known
corpora, and 4 exploiting one Master Key vulnerability each)
to all sandboxes which were available at the time of writing and
which use some form of dynamic analysis within a sandbox.
Sandboxes which were not available during our evaluation
were excluded, leaving ten sandboxes. Our evaluation is based
on whether the sandboxes detect the malicious behavior. The
results of our evaluation are outlined in Table III.
For every family of malware, we submitted two different
samples (separated by a “/” in Table III). Andrubis was able to
analyze every submitted sample, while the Nyleaker malware
was classified as benign. ForeSafe was able to successfully flag
every submitted sample as high risk malware. SandDroid was
unable to analyze any submitted sample. As for CopperDroid,
just one version of Geinimi and DroidKungFu were detected.
22http://nakedsecurity.sophos.com/2013/11/06/anatomy-of-a-file-format-problem-
yet-another-code-verification-bypass-in-android/
23http://bugs.python.org/issue14315
TABLE III. ANALYSIS RESULTS OF ONLINE SANDBOXES FOR TWO
SAMPLES PER MALWARE FAMILY (“”=DETECTED, “”=NOT DETECTED,
“-”=ANALYSIS ERROR).
Framework Obad Geinimi DroidKungFu Basebridge/
Nyleaker
Andrubis / / / /
CopperDroid - / - / - - / - / -
ForeSafe / / / /
Joe Sandbox Mobile / / / /
Mobile Sandbox - / - - / - - / - - / -
SandDroid - / - - / - - / - - / -
TraceDroid / / / /
VisualThreat / - / / /
TABLE IV. EVALUATION OF SANDBOXES WITH MASTER KEY SAMPLES:
”” INDICATES THAT THE SAMPLE WAS SUCCESSFULLY EXECUTED, ”-”
INDICATES, THAT THE SANDBOX WAS NOT ABLE TO EXECUTE THE
SAMPLE.
Framework Bug
8219321
Bug
9695860
Bug
9950697
Python
ZIP Bug
Andrubis - -
CopperDroid - - - -
ForeSafe
TraceDroid - -
VisualThreat -
For all other samples the analysis was aborted with an ”Instal-lation
Error”. Joe Sandbox Mobile as well as TraceDroid on the
contrary were able to analyze every version of every sample we
submitted. Joe Sandbox Mobile furthermore correctly flagged
every sample as malicious in the generated reports.
We also compiled an APK for each of the described Master
Key bugs and submitted them to our selection of sandboxes
to see if they could analyze malicious APKs exploiting those
vulnerabilities. We present the results in Table IV. As we will
discuss in Section V-C we were not able to get results from
all sandboxes for our evaluation in time, as several sandboxes
were out of service or did not produce any reports in a
reasonable amount of time.
C. Limitations of Existing Sandboxes
We observed that many sandboxes were not able to fully
analyze our samples, were prone to bugs or evasion techniques,
or were either no longer maintained or not publicly available

anymore at the time of writing. As some versions of the
Obad family evade DroidBox, Mobile Sandbox was not able
to analyze these. We verified this in a personal conversation
with the author of Mobile Sandbox. The other samples, namely
Geinimi, DroidKungFu, Basebridge and Nyleaker, are also
marked as errors in our table as there are currently over
300,000 samples in the queue to be analyzed by the dynamic
sandbox. Any submitted sample would thus be analyzed after
approximately 400 days, which is too long in our point of
view. Unfortunately, this number does not seem to decrease.
SandDroid accepts samples for analysis but is not able to
perform any analysis on the samples. After submitting the
samples and waiting for about two weeks, the samples are
either not in the report database or marked as still being
analyzed by the framework.
Some frameworks mentioned in Table II were excluded
from our evaluation: AASandbox, AppIntent, ANANAS,
SmartDroid as well as vetDroid were neither available as
source code nor as an online submission. We were not able to
run any samples for analysis within these frameworks. Because
AppsPlayground has no online submission possibility24 it was
out of scope and therefore also excluded from our evaluation.
D. Discussion
Every sandbox carries out a multitude of analysis tech-niques,
ranging from basic static to rigorous dynamic analysis
techniques. Due to the large number of available analysis
frameworks, on- as well as offline, it is impossible for a
user to determine which framework offers the most and
comprehensive set of features. This fact is substantiated by
the ongoing improvements in the mobile sector, like Google’s
new Android Runtime (ART) technology25, that will also
complicate static and dynamic analysis in the future due to
possibly pre-compiled applications.
Supporting these facts, there is no Swiss-army-knife-sandbox
that on one hand offers every possible feature and
on the other hand is readily available. Nevertheless, also on
mobile devices the cat-and-mouse-game between malware au-thors
and security researchers is continuously ongoing, as it has
been on personal computers for many years. One possibility to
deal with that problem could be one non-commercially driven
analysis framework that implements all static and dynamic
analysis techniques discussed in the literature so far, and is
fully maintained as well as continuously extended by research-ing
volunteers. Until this goal is reached, malware authors will
be ahead of defending researchers and industrial professionals.
We tried to contact every sandbox creator and notified them
of our findings.
VI. CONCLUSION
In this paper we conducted a comprehensive study on
detection methods of Android malware. First, we provided an
overview on current Android malware distribution techniques
and the motivation of malware authors. Secondly, we analyzed
available dynamic analysis platforms for Android and exam-ined
their interdependencies. An evaluation of ten sandboxes
24http://list.cs.northwestern.edu/mobile/
25http://source.android.com/devices/tech/dalvik/art.html
which are available as online services, using samples from four
real-life Android malware families, showed that detection rates
vary. In addition we found that popular dynamic sandboxes are
susceptible to well known vulnerabilities like the Master Key
vulnerabilities, which could be potentially misused by Android
malware to thwart analysis. In conclusion, our findings show
that while current dynamic sandboxes are valuable for aca-demic
research to analyze Android malware, they can not be
considered an effective defense against malware on their own.
Furthermore, interdependencies between platforms caused by
code reuse can lead to challenges in detecting malware that
targets specific platform limitations, as all analysis platforms
which share the same code base are affected. Thus, while a
number of different analysis sandboxes exist, diversity among
them is low.
ACKNOWLEDGEMENTS
The research was funded by COMET K1, FFG - Aus-trian
Research Promotion Agency. The research leading to
these results has also received funding from the European
Union Seventh Framework Programme (FP7/2007-2013) under
grant agreement no 257007. Moreover this work has been
carried out within the scope of u’smile, the Josef Ressel
Center for User-Friendly Secure Mobile Environments. We
gratefully acknowledge funding and support by the Christian
Doppler Gesellschaft, A1 Telekom Austria AG, Drei-Banken-
EDV GmbH, LG Nexera Business Solutions AG, and NXP
Semiconductors Austria GmbH.
REFERENCES
[1] Canalys, “Over 1 billion Android-based smart phones to
ship in 2017,” http://www.canalys.com/newsroom/over-
1-billion-android-based-smart-phones-ship-2017, June
2013.
[2] R. Llamas, R. Reith, and M. Shirer, “Apple Cedes
Market Share in Smartphone Operating System Market
as Android Surges and Windows Phone Gains, Accord-ing
to IDC,” http://www.idc.com/getdoc.jsp?containerId=
prUS24257413, August 2013.
[3] E. Kalige and D. Burkey, “A Case Study of Eurograbber:
How 36 Million Euros was Stolen via Malware,”
https://www.checkpoint.com/products/downloads/
whitepapers/Eurograbber White Paper.pdf, December
2012.
[4] M. Lennon, “Resilient ’SMSZombie’ Infects 500,000
Android Users in China,” http://www.securityweek.
com/resilient-smszombie-infects-500000-android-users-china,
August 2012.
[5] C. Warren, “Google Play Hits 1 Million Apps,” http:
//mashable.com/2013/07/24/google-play-1-million, July
2013.
[6] AppBrain, “Number of available Android applications,”
http://www.appbrain.com/stats/number-of-android-apps.
[7] H. Lockheimer, “Android and Security,”
http://googlemobile.blogspot.com/2012/02/android-and-
security.html, February 2012.
[8] X. Jiang, “An Evaluation of the Application (”App”)
Verification Service in Android 4.2,” http://www.cs.ncsu.
edu/faculty/jiang/appverify, Dec. 2012.
[9] J. Oberheide and C. Miller, “Dissecting the Android
Bouncer,” in SummerCon, 2012.

[10] T. Bl¨asing, L. Batyuk, A.-D. Schmidt, S. A. Camtepe, and
S. Albayrak, “An Android Application Sandbox System
for Suspicious Software Detection,” in Proceedings of the
5th International Conference on Malicious and Unwanted
Software (MALWARE), 2010.
[11] C. Willems, T. Holz, and F. Freiling, “Toward Automated
Dynamic Malware Analysis Using CWSandbox,” IEEE
Security and Privacy, vol. 5, no. 2, 2007.
[12] U. Bayer, C. Kruegel, and E. Kirda, “TTAnalyze: A
Tool for Analyzing Malware,” in Proceedings of the
15th European Institute for Computer Antivirus Research
(EICAR) Annual Conference, 2006.
[13] M. Egele, T. Scholte, E. Kirda, and C. Kruegel, “A
Survey on Automated Dynamic Malware Analysis Tech-niques
and Tools,” ACM Computing Surveys Journal,
vol. 44, no. 2, 2012.
[14] D. Maslennikov, “First SMS Trojan for Android,”
https://www.securelist.com/en/blog/2254/First SMS
Trojan for Android, August 2010.
[15] Alcatel-Lucent, “Kindsight Security Labs Malware Re-port
- Q2 2013,” http://www.kindsight.net/sites/default/
files/Kindsight-Q2-2013-Malware-Report.pdf, Jul. 2013.
[16] Trend Micro, “TrendLabs 2Q 2013 Security
Roundup,” http://www.trendmicro.com/cloud-content/
us/pdfs/security-intelligence/reports/rpt-2q-
2013-trendlabs-security-roundup.pdf, August 2013.
[17] D. Maslennikov, “Zeus-in-the-Mobile — Facts and The-ories,”
http://www.securelist.com/en/analysis/204792194,
October 2011.
[18] Juniper Networks, “Juniper Networks Third Annual
Mobile Threats Report,” http://www.juniper.net/us/en/
local/pdf/additional-resources/jnpr-2012-mobile-threats-report.
pdf, June 2013.
[19] W. Zhou, Y. Zhou, X. Jiang, and P. Ning, “Detecting
Repackaged Smartphone Applications in Third-Party An-droid
Marketplaces,” in Proceedings of the 2nd ACM
Conference on Data and Application Security and Pri-vacy
(CODASPY), 2012.
[20] A. Bibat, “GGTracker Malware Hides As Android
Market,” http://www.androidauthority.com/ggtracker-malware-
hides-as-android-market-17281/, June 2011.
[21] H. Yao and D. Shin, “Towards Preventing QR Code
Based Attacks on Android Phone Using Security Warn-ings,”
in Proceedings of the 8th ACM Symposium on
Information, Computer and Communications Security
(ASIACCS), 2013.
[22] Google, “Android Developers: Platform Versions,” http:
//developer.android.com/about/dashboards/index.html.
[23] “RageAgainstTheCage,” http://thesnkchrmr.wordpress.
com/2011/03/24/rageagainstthecage/, 2011.
[24] “udev Exploit (exploid),” http://thesnkchrmr.wordpress.
com/2011/03/27/udev-exploit-exploid, 2011.
[25] “zergRush,” http://github.com/revolutionary/zergRush.
[26] Y. Zhou and X. Jiang, “Dissecting Android Malware:
Characterization and Evolution,” in Proceedings of the
33rd Annual IEEE Symposium on Security and Privacy
(SP), 2012.
[27] “Contagio Mobile,” http://contagiominidump.blogspot.
com.
[28] “VirusShare,” http://www.virusshare.com.
[29] “VirusTotal,” http://www.virustotal.com.
[30] F. Maggi, A. Valdi, and S. Zanero, “AndroTotal: A Flex-ible,
Scalable Toolbox and Service for Testing Mobile
Malware Detectors,” in Proceedings of the 3rd ACM
Workshop on Security and Privacy in Smartphones and
Mobile Devices (SPSM), 2013.
[31] A. Desnos and G. Gueguen, “Android: From Reversing
to Decompilation,” in Black Hat Abu Dhabi, Dec. 2011.
[32] W. Enck, D. Octeau, P. McDaniel, and S. Chaudhuri, “A
Study of Android Application Security,” in Proceedings
of the 20th USENIX Security Symposium, 2011.
[33] D. Octeau, W. Enck, and P. McDaniel, “The ded Decom-piler,”
Network and Security Research Center, Depart-ment
of Computer Science and Engineering, Pennsylva-nia
State University, University Park, PA, USA, Tech.
Rep. NAS-TR-0140-2010, Sep. 2010.
[34] T. Garfinkel and M. Rosenblum, “A Virtual Machine In-trospection
Based Architecture for Intrusion Detection,”
in Proceedings of the 10th Annual Network Distributed
System Security Symposium (NDSS), 2003.
[35] T. Raffetseder, C. Kruegel, and E. Kirda, “Detecting Sys-tem
Emulators,” in Proceedings of the 10th Information
Security Conference (ISC), 2007.
[36] W. Enck, P. Gilbert, B.-G. Chunn, L. P. Cox, J. Jung,
P. McDaniel, and A. N. Sheth, “TaintDroid: An
Information-Flow Tracking System for Realtime Privacy
Monitoring on Smartphones,” in Proceedings of the 9th
USENIX Symposium on Operating Systems Design and
Implementation (OSDI), 2010.
[37] G. Sarwar, O. Mehani, R. Boreli, and M. A. Kaafar, “On
the Effectiveness of Dynamic Taint Analysis for Pro-tecting
Against Private Information Leaks on Android-based
Devices,” in Proceedings of the 10th International
Conference on Security and Cryptography (SECRYPT),
2013.
[38] Y. Zhang, M. Yang, B. Xu, Z. Yang, G. Gu, P. Ning,
X. Wang, and B. Zang, “Vetting Undesirable Behaviors
in Android Apps with Permission Use Analysis,” in
Proceedings of the 20th ACM Conference on Computer
and Communications Security (CCS), 2013.
[39] “DroidBox: An Android Application Sandbox for Dy-namic
Analysis,” https://code.google.com/p/droidbox/.
[40] T. Eder, M. Rodler, D. Vymazal, and M. Zeilinger,
“ANANAS - A Framework For Analyzing Android Ap-plications,”
in Proceedings on the 1st International Work-shop
on Emerging Cyberthreats and Countermeasures
(ECTCM), 2013.
[41] L. K. Yan and H. Yin, “DroidScope: Seamlessly Recon-structing
the OS and Dalvik Semantic Views for Dynamic
Android Malware Analysis,” in Proceedings of the 21st
USENIX Security Symposium, 2012.
[42] L. Weichselbaum, M. Neugschwandtner, M. Lindor-fer,
Y. Fratantonio, V. van der Veen, and C. Platzer,
“Andrubis: Android Malware Under The Magnifying
Glass,” Vienna University of Technology, Tech. Rep. TR-ISECLAB-
0414-001, 2014.
[43] V. van der Veen, “Dynamic Analysis of Android Mal-ware,”
Internet Web Technology Master thesis, VU
University Amsterdam, 2013.
[44] M. Spreitzenbarth, F. Freiling, F. Echtler, T. Schreck, and
J. Hoffmann, “Mobile-Sandbox: Having a Deeper Look
into Android Applications,” in Proceedings of the 28th
Annual ACM Symposium on Applied Computing (SAC),
2013.

[45] A. Reina, A. Fattori, and L. Cavallaro, “A System Call-
Centric Analysis and Stimulation Technique to Auto-matically
Reconstruct Android Malware Behaviors,” in
Proceedings of the 6th European Workshop on System
Security (EUROSEC), 2013.
[46] C. Zheng, S. Zhu, S. Dai, G. Gu, X. Gong, X. Han,
and W. Zou, “SmartDroid: An Automatic System for
Revealing UI-based Trigger Conditions in Android Ap-plications,”
in Proceedings of the 2nd ACM Workshop on
Security and Privacy in Smartphones and Mobile Devices
(SPSM), 2012.
[47] V. Rastogi, Y. Chen, and W. Enck, “AppsPlayground: Au-tomatic
Security Analysis of Smartphone Applications,”
in Proceedings of the 3rd ACM Conference on Data and
Application Security and Privacy (CODASPY), 2013.
[48] Z. Yang, M. Yang, Y. Zhang, G. Gu, P. Ning, and
X. S. Wang, “AppIntent: Analyzing Sensitive Data Trans-mission
in Android for Privacy Leakage Detection,” in
Proceedings of the 20th ACM Conference on Computer
and Communications Security (CCS), 2013.
[49] D. Barrera, J. Clark, D. McCarney, and P. C. van
Oorschot, “Understanding and Improving App Installa-tion
Security Mechanisms Through Empirical Analysis of
Android,” in Proceedings of the 2nd ACM Workshop on
Security and Privacy in Smartphones and Mobile Devices
(SPSM), 2012.
[50] P. Traynor, M. Lin, M. Ongtang, V. Rao, T. Jaeger, P. Mc-
Daniel, and T. La Porta, “On Cellular Botnets: Measuring
the Impact of Malicious Devices on a Cellular Network
Core,” in Proceedings of the 16th ACM Conference on
Computer and Communications Security (CCS), 2009.
[51] M. G. Kang, H. Yin, S. Hanna, S. McCamant, and
D. Song, “Emulating Emulation-Resistant Malware,” in
Proceedings of the 2nd Workshop on Virtual Machine
Security (VMSec), 2009.
[52] D. Balzarotti, M. Cova, C. Karlberger, E. Kirda,
C. Kruegel, and G. Vigna, “Efficient Detection of Split
Personalities in Malware,” in Proceedings of the 17th An-nual
Network Distributed System Security Symposium
(NDSS), 2010.
[53] M. Lindorfer, C. Kolbitsch, and P. Milani Comparetti,
“Detecting Environment-Sensitive Malware,” in Proceed-ings
of the 14th International Conference on Recent
Advances in Intrusion Detection (RAID), 2011.
[54] F. Matenaar and P. Schulz, “Detecting Android Sand-boxes,”
http://dexlabs.org/blog/btdetect, August 2012.
[55] P. Ferrie, “Attacks on More Virtual Machine Emulators,”
Symantec Research White Paper, Tech. Rep., 2007.
[56] C. Kolbitsch, E. Kirda, and C. Kruegel, “The Power of
Procrastination: Detection and Mitigation of Execution-
Stalling Malicious Code,” in Proceedigns of the 18th
ACM Conference on Computer and Communications
Security (CCS), 2011.
[57] R. Unucheck, “The most sophisticated Android Tro-jan,”
http://www.securelist.com/en/blog/8106/The most
sophisticated Android Trojan, June 2013.
[58] Comodo Malware Analysis Team, “Android OBAD
- Technical Analysis Paper,” http://www.comodo.com/
resources/Android OBAD Tech Reportv3.pdf, 2013.
[59] Z. Ashraf, “DIY: Android Malware Analysis – Taking
apart OBAD (part 1),” http://securityintelligence.com/
diy-android-malware-analysis-taking-apart-obad-part-1/,
Oct 2013.
[60] X. Jiang, “Security Alert: New Sophisticated An-droid
Malware DroidKungFu Found in Alternative
Chinese App Markets,” http://www.csc.ncsu.edu/faculty/
jiang/DroidKungFu.html, 2011.
[61] Symantec Security Response, “First Malicious Use
of ’Master Key’ Android Vulnerability Discovered,”
http://www.symantec.com/connect/blogs/first-malicious-use-
master-key-android-vulnerability-discovered, July
2013.

Enter Sandbox: Android Sandbox Comparison

More Related Content

What's hot (20)

Viewers also liked (6)

Similar to Enter Sandbox: Android Sandbox Comparison (20)

Recently uploaded (20)

Enter Sandbox: Android Sandbox Comparison