An Old Bot’s Nasty New Tricks: Exploring Qbot’s Latest Attack Methods

August 27, 2020

Research By: Alex Ilgayev


The notorious banking trojan Qbot has been in business for more than a decade. The malware, which has also been dubbed Qakbot and Pinkslipbot, was discovered in 2008 and is known for collecting browsing data and stealing banking credentials and other financial information from victims. It is highly structured, multi-layered, and is being continuously developed with new features to extend its capabilities. .These new ‘tricks’ mean that despite its age, Qbot is still a dangerous and persistent threat to organizations.  It has become the malware equivalent of a Swiss Army knife, capable of:

  • Stealing information from infected machines, including passwords, emails, credit card details and more.
  • Installing other malware on infected machines, including ransomware 
  • Allowing the Bot controller to connect to the victim’s computer (even when the victim is logged in) to make banking transactions from the victim’s IP address
  • Hijacking users’ legitimate email threads from their Outlook client and using those threads to try and infect other users’ PCs

A prominent campaign using QBot ran from March to the end of June this year.   We assumed that the campaign was stopped to allow those behind QBot to conduct further malware development, but we did not imagine that it would return so quickly.

Towards the end of July, one of today’s most serious cyber threats, the Emotet Trojan, returned to full activity and launched multiple malspam campaigns, impacting 5% of organizations globally. Some of these campaigns included installing an updated version of Qbot on victims’ PCs. A few days later, we identified a newer Qbot sample dropped by latest Emotet campaign, which led us discovering a renewed command and control infrastructure and brand new malware techniques distributed through Emotet’s infection process.

If that wasn’t enough for us, Qbot’s malspam campaign resumed earlier in August, spreading globally and infecting new targets. One of Qbot’s new tricks is particularly nasty, as once a machine is infected, it activates a special ‘email collector module’ which extracts all email threads from the victim’s Outlook client, and uploads it to a hardcoded remote server. These stolen emails are then utilized for future malspam campaigns, making it easier for users to be tricked into clicking on infected attachments because the spam email appears to continue an existing legitimate email conversation.  Check Point’s researchers have seen examples of targeted, hijacked email threads with subjects related to Covid-19, tax payment reminders, and job recruitments.

Based on our visibility, most of the attacks were made against US- and Europe-based organizations as we can see in Figure 1.

Figure 1 – Attacked organizations by country

Among these, the most targeted industries were in the government, military, and manufacturing sectors.

Figure 2 – Attacked Organizations by industry

After a thorough analysis of these new QBot samples, we will share our knowledge and insights about the following topics:

  • Qbot’s infection process – hijacked email threads and VBS downloaders.
  • Its payload functionality and version break down.
  • C&C communication protocol and module fetching.
  • How a victim becomes a potential bot-proxy, and various methods that the Proxy module exposes.

Infection Chain

QBot’s infection chain is described in the following flow-chart and will be discussed in the next sections:

Figure 3 – Infection chain diagram

Malicious Email

The initial infection chain starts by sending specially crafted emails to the target organizations. The method is less sophisticated than spear-phishing techniques but has additional attributes which add to its credibility. One of these is called “Hijacked Email Threads” – capturing archived email conversations and replying to the sender with the malicious content. Those conversations could be captured using Qbot’s Email Collector module which we will describe later.

We can see in Figure 4, Figure 5, and Figure 6, examples of such methods from samples submitted by @malware_traffic in their blog:

Figure 4 – Example of COVID-19 related email thread

Figure 5 – Example of email thread for tax payment reminder

Figure 6 – Example of email thread for job recruitment

Each of these emails contains a URL to a ZIP with a malicious VBS – Visual Basic Script file.

During our tracking of the malspam campaign, we have seen hundreds of different URLs for malicious ZIP dropping when most of them were compromised WordPress sites.

VBS Infection

The VBS based infection method is rather new for the malware, and is being used since April 2020. In previous campaigns, the infection chain started with a Word document containing malicious macros.

While the previous macros had simple obfuscation and string decoding, the VBS file contains several more advanced methods:

File Size – The file size is larger than 35MB, padded with NULL bytes. Big files are usually dismissed by various sandboxes due to performance limitations.

Sleep Timer – The script delays its execution by calling the Sleep API. This is another method for avoiding sandboxes.

Obfuscation – The script contains multiple obfuscation methods such as those described in Figure 7.

Figure 7 – VBS obfuscation methods

Encryption – The VBS file downloads the Qbot payload from one of 6 possible hardcoded encrypted URLs. These URLs are encrypted by a custom XOR encryption 3 times with different keys that are built dynamically. We created an extraction script that can be accessed in Appendix B.

In order to support detection and hunting for additional malicious VBS files, We wrote a YARA rule which can be observed in Appendix A.

Similar to the old infection method, the VBS file downloads and executes the Qbot payload.

Qbot Payload

Version Analysis

In the course of our analysis, Qbot’s operators frequently upgraded its versions and encouraged us to track and analyze the changes in each and every version. The fact that the developers left a version tag marked in the samples, allowed us to perform this analysis easier.

For example, let’s have a look at the version tag as it shown in the unpacked sample below:

Figure 8 – Sample’s major and minor version

From that, we can deduct that the initial payload version is 325/5, while the main payload version is 325/7. (The major version is read as a hex value)

Over the last few months, we tracked the different versions of Qbot and identified some of the differences in each version, as can be seen in the following table.

Major Minor Payload Minor Version Timestamp  Notes
324 44 8 Jan 22, 2020 First version seen for major version 324.
324 353 53 Mar 3, 2020  
324 375 65 Mar 13, 2020  
324 379 70 Mar 20, 2020 Added command 35 supporting hVNC module.
324 383 74 Apr 1, 2020  
324 385 75 Apr 1, 2020  
324 388 79 Apr 8, 2020 Added command 10 – terminate process by name.
324 390 127 Apr 10, 2020  
324 393 136 Apr 29, 2020 JS Updater resource is no longer included. JS Update commands has been respectively adjusted.
324 399 141 May 7, 2020 Added long list of blacklisted analysis programs part of anti-VM method.
324 401 142 May 28, 2020  
325 5 7 July 29, 2020 Introduced new anti-analysis techniques.
Added anti-VM checks on server-side.
325 7 13 July 31, 2020  
325 8 14 August 3, 2020  
325 35 42 August 7, 2020  
325 37 43 August 11, 2020 Last known version up to the writing of this article.

The date mentioned for each sample is based on the executable compilation time attribute. That field can be changed via timestomping, but we suspect that it wasn’t forged in these cases.

We also tracked the timestamps of the main payload, and seen that the compilation time was consistently minutes apart from the initial payload’s:

Figure 9 – Comparing sample-to-payload compilation time

Decryption Schemes

The malware implements several encryption schemes to conceal its functionality and data from the victims, and Anti-Virus vendors. In order to successfully analyze the malware and its components, we had to automate the decryption process for all the variants.

The following table shows the different decryption and decoding methods:

Encrypted Data Algorithm Key Source
Network Data Base64 + RC4 KEY = SHA1(ENCRYPTED[0:16] + "jHxastDcds)oMc=jvh7wdUhxcsdt2")
Payload (Resource “307”) RC4 + Custom Compression KEY = ENCRYPTED[0:20]
Javascript Updater File RC4 + Custom Compression KEY = ENCRYPTED[0:20]
Bot List (Resource “311”) and Initial Configuration (Resource “308”) RC4 KEY = ENCRYPTED[0:20]
Configuration “.dat” File, Web-Inject File, Hooking Module RC4 KEY = SHA1(EXE_NAME)
Stolen Information “.dll” File RC4 srand(CRC32(BOT_ID))

Initial Payload

Qbot’s initial payload has been covered extensively by fellow malware researchers. The latest versions have implemented several typical malware components to reduce its visibility and toughen its analysis:

Packer – The executable has been reconstructed using a packer.

Random Directory Name – Creating a working directory with randomized directory and file names to avoid file signatures. Directory location is %APPDATA%\Microsoft. Working directory example can be observed at Figure 11.

String Encryption – Containing encrypted strings using XOR encryption (applies also to other modules).

Dynamic Import Table – Import table built dynamically based on encrypted strings (applies also to other modules).

Anti-VM and Anti-Debug Techniques:

  • Latest versions are looking for VM-related artifacts on server-side. victim computer configuration is being enumerated and sent to the C2. Based on that information, the server decides whether is safe to “push” modules to the victim.
  • Looking for “VMWare” port existence
  • Looking for VM and analysis related processes. The latest versions also adds a long list of blacklisted analysis programs:

Fiddler.exe;samp1e.exe;sample.exe;runsample.exe;lordpe.exe;regshot.exe;Autoruns.exe;dsniff.exe;VBoxTray.exe;HashMyFiles.exe;ProcessHacker.exe;Procmon.exe;Procmon64.exe;netmon.exe;vmtoolsd.exe;vm3dservice.exe;VGAuthService.exe;pr0c3xp.exe;ProcessHacker.exe; CFF Explorer.exe;dumpcap.exe;Wireshark.exe;idaq.exe;idaq64.exe;TPAutoConnect.exe;ResourceHacker.exe;vmacthlp.exe;OLLYDBG.EXE;windbg.exe;bds-vision-agent-nai.exe;bds-vision-apis.exe; bds-vision-agent-app.exe;MultiAnalysis_v1.0.294.exe;x32dbg.exe;VBoxTray.exe;VBoxService.exe;Tcpview.exe

  • Looking for VM related device drivers. Examples:

Figure 10 – Device driver Anti-VM technique

  • Looking for a VM through CPUID instruction
  • Forcing exceptions to check if debugger is present
  • Checking for sandbox signatures

Persistence – Achieving persistence through registry values and task scheduler.

Whenever the malware decides it is safe to run on the targeted system, it decrypts its resource “307” as explained above, injects it into newly created process explorer.exe, calls a loader procedure which loads the DLL and calls the DllEntryPoint of the main payload.

Main Payload

The main payload has multiple roles:

  • Creating and maintaining the configuration of the malware.
  • Creating and maintaining the messaging mechanisms – Named pipes, events, and custom Windows messages.
  • Installing and managing new modules – new feature.
  • Creating and maintaining a proper communication channel with the C&C server.
  • Executing commands through a custom thread queue mechanism.

The payload has several more internal modules that we won’t elaborate on in this article such as – lateral movement capabilities, certificates harvesting, spam bot, and more.

The malware constructs its configuration out of several embedded resources that are unpacked and decrypted on runtime. The resources are:

  • “308” – Initial configuration data.
  • “311” – List of 150 bots IP addresses and ports for building a communication tunnel.

The working directory, as we can see in Figure 11, is an important part of the Qbot’s functionality, and is also used as a synchronization method between the modules.

Figure 11 – Example for Qbot’s working directory

Qbot’s configuration files (end with .dat) and stolen information files (end with .dll) are the most crucial. These files are accessed and loaded by all of its modules.

Figure 12 – Configuration file

Figure 13 – Stolen information file

One of the questions we were asking ourselves at this point of the research was where can we find the real “banking” logic. Older versions of Qbot contained multiple malicious modules as embedded resources, but recent versions were rather “clean”.

To understand that, we had to dive deeper into the communication protocol and find methods to fetch the malicious modules.

Communication Module

Resource “311”, as we stated previously, contains a list of 150 IP addresses of other bots for the victim to communicate with. Each of these bots will forward the traffic to the real C&C server or to a second-tier proxy as we will show later. This method is an efficient way of hiding the C&C IP address.

All the following messages are sent via POST method to the next URL: https://<BOT_IP>:<BOT_PORT>/t3 and are encrypted with a random initialization value. To make it easier understanding the logic, we will show only the decrypted network data.

The C&C communication data is sent in JSON format, where each property is identified by unique numerical ID. As we can see later in sample messages, most important JSON property is its message code which holds the key 8. We were able to map the next unique message codes:

Victim → C&C:

1 – Request the next command from C&C.

2 – Ack for a command given by C&C.

4 – Computer configuration and process enumeration.

7 – Report of stolen information.

8 – Open ports message.

9 – Keep-alive message.

C&C → Victim:

5 – Server Ack.

6 – Command to execute.

The program holds two parallel networking loops – Keep-alive and report session, and Command Execution Session.

Keep-alive and Report Session

This session is pretty simple. The program alternates between keep-alive message to stolen information report message. For each such message, it will receive a server ack. These messages will look as follows:

Keep-alive Message

// Victim -> C&C
    "8": 9, // MSG code
    "1": 17, // Network protocol version
    "2": "powqdc619830" // Victim BOT ID

Report Stolen Information Message

Takes the encrypted .dll file of the stolen information, and sends it.

// Victim -> C&C
    "8": 7, // MSG code
    "1": 17, // Network protocol version
    "2": "powqdc619830", // Victim BOT ID
    "3": "spx145", // Bot group
    "6": 223,
    "7": 4763,
    "36": "617c...icR67==" // Base64 encoded and encrypted information

Keep-alive and Report Response

// C&C -> Victim
    "8": 5, // MSG code
    "16": 270544960, // Victim IP address
    "39": "mzJzbJU", // Random data
    "38": 1

Command Execution Session

The malware will request new commands periodically and execute them according to the following command table. The table contains the appropriate command ID and its handler.

Figure 14 – Qbot’s command table

The command request message will have the next structure:

    "8": 1, // MSG code
    "1": 17, // Network protocol version
    "2": "powqdc619830", // Victim BOT ID
    "3": "b", // Bot group
    "4": 804, // Payload major version
    "5": 141,   // Payload minor vesion
    "10": "1582872269", // Timestamp
    "6": 6210,
    "7": 6278,
    "14": "U3HphEKFiQcKFFe0LUVZNDO9vsJ9zdEf09"

A typical response would look like the following:

    "8": 6, // MSG code
    "15": "...",
    "16": 270544960, // Victim IP address
    "18": 252,
    "19": 31, // command ID to execute
    "20": ["TVqQAAM...="], // command payload
    "39": "<RANDOM_STRING>"  // Random data

Module Fetching

During the research we were able to map several downloaded modules, some of which were newly added as we could see in version break down.

We noticed that whenever a new Bot ID is being “registered” by the C&C server, on the next command request it will receive the next modules to download and install:

Executable Update – Updates the current executable with a newer version or newer bot list. The C&C periodically pushes updates to all of its victims.

Email Collector Module – Extracts all e-mail threads from the victim’s Outlook client by using MAPI32.dll API, and uploads it to a hardcoded remote server. These stolen e-mails will be utilized for the malspam to come later.

Figure 15 – Email collector module

Hooking Module – The module injects itself to all running processes, and hooks relevant API functions. Sample hooking table:

Figure 16 – Hooking module

Web-Inject File – The file provides the injector module with a list of websites and JavaScript code that will be injected if the victim visits any of these websites.

We can see the results of visiting one of the actor’s targets – Chase Bank.

Figure 17 – HTML source code example for an injected target

Password Grabber Module – a large module that downloads Mimikatz and tries to harvest passwords.

Figure 18 – Password grabber module

hVNC Plugin – Allows controlling the victim machine through a remote VNC connection. That is, an external operator can perform bank transactions without the user’s knowledge, even while he is logged into his computer. The module shares a high percentage of code with similar modules like TrickBot’s hVNC.

Figure 19 – Hidden VNC plugin

JS Updater Loader – Decrypts and writes a Javascript updater script. Until recently, the script came as an encrypted resource inside the payload. Because the script contains encrypted hardcoded URLs, the new method makes it easier for the operator to push updated domains to the victims.

We wrote a Python script for ease URL extraction from a given script which can be observed in Appendix C.

Figure 20 – JS updater script example

Cookie Grabber Module – targets popular browsers: IE, Edge, Chrome, and Firefox.

Figure 21 – Cookie grabber module

We can identify these modules through a traffic capture program:

Figure 22 – Downloaded modules in Fiddler

Once the victim has been infected, their computer is compromised, and they are also a potential threat to other computers in the local network because of Qbot’s lateral movement capabilities. The malware then checks whether the victim can also be a potential bot as part of Qbot’s infrastructure.

From a Victim to a Bot

McAfee has published a great article 3 years ago in which they covered important details regarding the bot proxy module. To understand the complete infection chain process we felt there is more to discover regarding that module, and ways of fetching it.

To reach that goal, we started analyzing Qbot’s efforts of converting an innocent victim machine into an active bot, and being part of the C&C infrastructure. To do so, the malware does the following:

  • Execute shell commands to allow incoming connections in the host firewall.
  • Sending crafted UPnP commands to allow port forwarding.
  • Whenever it creates the opened ports list, the program verifies whether the incoming connection is really allowed by sending the next message to a remote bot and waiting for a connection.
    • URL – https://<BOT_IP>:<BOT_PORT>/bot_serv
    • Sample payload:
      • cmd=1&msg=J3zeJrBLh2sGU4ql0EIr9MncSBCnK&ports=443,995,993,465,990,22,2222,2078,2083,2087,1194,8443,20,21,53,80,3389,6881,6882,6883,32100,32101,32102,32103,50000,50001,50002,50003,50010,61200,61201,61202
  • The remote bot tries to connect using the specified ports to the victim. If the victim receives the data they expected (msg variable), then it’s a sign of a successful incoming connection.
  • Remove the port listening.

When the program finishes verifying its potential ports, it forms message code 8 and sends it to the C&C server:

    "8": 8, // MSG code
    "1": 17, // Network protocol version
    "2": "jnugfv895664", // Victim BOT ID
    "4": 3,
    "5": 111,
    "55": 270544960, // External IP of the potential bot
    "56": [443, 995, 993, 465, 990, 22, 2222, 2078, 2083, 2087, 1194, 8443, 20, 21, 53, 80, 3389, 6881, 6882, 6883, 32100, 32101, 32102, 32103, 50000, 50001, 50002, 50003, 50010, 61200, 61201, 61202] // Potential ports

When the program does this specific process, we could observe that on the next command execution request, we will receive a proxy module installation with the relevant port to listen:

    "8": 6, // MSG code
    "19": 25, // command ID
    "20": ["TVqQAAM...=", "prt=443", "n=jnugfv895664"], // command payload

We can visualize the process with the next diagram, and observe it through a traffic capture program:

Figure 23 – Network flow for proxy module download

Figure 24 – Downloaded proxy module in Fiddler

Analyzing Proxy Module

The proxy module is loaded by rundll32.exe, and copied into its working folder – C:\ProgramData\FilesystemMonitor\. If it is given SYSTEM privileges, it creates a new service named fsmon, otherwise updates CurrentVersion\Run registry value.

Most of the module’s codebase is taken from the following open-source libraries:

  • libcurl 7.47.1 for HTTP requests.
  • OpenSSL 1.0.2r 26 Feb 2019 – Used for certificate creation, and signature validation.
  • miniupnp – For port opening.

It also contains 3 hard-coded IP addresses of the second-tier proxy server.

The module hasn’t changed a lot since McAfee’s publication 3 years ago. The changes we could find were:

  • The service name, description, working folder, window name, and executable name has been changed. For example, the service name was changed from hwmon to fsmon.
  • OpenSSL version has been upgraded from 1.0.2f to 1.0.2r.
  • Updated Tier 2 Proxy servers.

One rather interesting feature of the proxy module is its control API. The threat group behind Qbot has developed a control API to the proxy which is independent of the malicious payload update mechanism. That API is also unique, mainly because it receives control messages by pushing and not by pulling, which could expose the bots to external actors’ control.

The protocol is rather simple and can be observed in the next diagram:

Figure 25 – Network flow for proxy module control API

The signature is being verified against the hardcoded public key of the actor. Hence, unless we possess the private key, the protocol is extremely hard to break.


This article analyzes two aspects of the threat – the campaign that leads to the infection of the victim, and the complex multi-layered malware which is constantly evolving. The article also covers several miscellaneous topics regarding its version history in the past year, decryption methods, communication samples, proxy server control API, and more.

These days Qbot is much more dangerous than it was previously – it has active malspam campaign which infects organizations, and it manages to use a “3rd party” infection infrastructure like Emotet’s to spread the threat even further. It seems like the threat group behind Qbot is evolving its techniques through the years, and Check Point Research hopes that the information in this article will help the researchers around the globe to mitigate and potentially stop Qbot’s activity.

Check Point SandBlast Agent protects against such attacks, and is capable of preventing them from the very first step.


Many Qbot and VBS samples were analyzed during the research. We’re attaching the recent samples and modules from 22/06/2020.


9001DF2C853B4BA118433DD83C17617E7AA368B1 – VBS Dropper
449F2B10320115E98B182204A4376DDC669E1369 – Qbot Sample SPX145
F85A63CB462B8FD60DA35807C63CD13226907901 – Mail Collector Module Loader [Decrypted]
B4BC69FF502AECB4BBC2FB9A3DFC0CA8CF99BA9E – Javascript Updater Loader [Decrypted]
1AAA14A50C3C3F65269265C30D8AA05AD8695B1B – Javascript Updater [Decrypted]
577522512506487C63A372BBDA77BE966C23CBD1 – Hooking Module Loader [Decrypted]
75107AEE398EED78532652B462B77AE6FB576198 – Cookie Grabber Module [Decrypted]
674685F3EC24C72458EDC11CF4F135E445B4185B – Password Grabber Module [Decrpyted]
BECD8F2D6289B51981F07D5FF52916104D764DD5 – hVNC Module [Decrpyted]
18E8971B2DE8EA3F8BB7E1462E414DA936425D4E – Proxy Module Loader [Decrypted]
4C96D2BCE0E12F8591999D4E00498BCDB8A116DE – Proxy Module

Domains and IPs



VBS Dropper URLs


Web-Inject URLs


Mail Collector Remote Server


Mimikatz URL Download


Tier 2 Proxy Servers

Javascript Updater URLs


Bot List

Other IOC

Proxy Service Name


Proxy Service Display name

Filesystem Monitor

Proxy File Paths


Proxy Executable Command Line

C:\Windows\SysWOW64\rundll32.exe "C:\ProgramData\FilesystemMonitor\fsmonitor.dll",FsMonServerMainNT
C:\Windows\SysWOW64\rundll32.exe "C:\ProgramData\FilesystemMonitor\fsmonitor.dll",#1

Proxy RSA Public Key



Appendix A: YARA Rule for VBS Hunting

rule qbot_vbs
        description = "Catches QBot VBS files"
        author = "Alex Ilgayev"
        date = "2020-06-07"
        $s3 = "ms.Send"
        $s4 = "for i=1 to 6"
        $s5 = "if ms.readyState = 4 Then"
        $s6 = "if len(ms.responseBody) <> 0 then"
        $s7 = /if left\(ms.responseText, \w*?\) = \"MZ\" then/

        filesize > 20MB and $s3 and $s4 and $s5 and $s6 and $s7

Appendix B: VBS URL Extraction Script

"""Qbot VBS URL extractor and de-obfuscator.

This script is for research purposes, and far from production ready (missing exception handling and more).

import re
import os
import sys

def remove_additions(lines):
    """Removes stub calculations. 
    IZLmoJg = 277 + 15 + 23 + 468 - 345 - 18 - 471 - 15 + 617
    Will be replaced with:
    IZLmoJg = 551

        lines (list): List of lines.

        list: List of modified lines.
    pattern = r'(([0-9]{1,15} [\+,\-] )+[0-9]{1,15})'
    new_file  = []

    for line in lines:
        line = line.strip()

        res =, line)
        if res:
            new_line = re.sub(pattern, lambda x:str(eval(, line)
            new_line = line
    return new_file

def remove_chr(lines):
    """Replaces "chr(*)" with their respective characters.

        lines (list): List of lines.

        list: List of modified lines.
    pattern = r'[c,C]hr\((\d?\d?\d?)\)'
    new_file  = []

    for line in lines:
        line = line.strip()

        res =, line)
        if res:
            new_line = re.sub(pattern, lambda match: '\"' + str(chr(int( + '\"', line)
            new_line = line
    return new_file

def remove_replace(lines):
    """Replaces "replace(*)" with its respective string.

        lines (list): List of lines.

        list: List of modified lines.
    pattern = r'replace\(\"(.*)\"\, \"(.*)\"\, \"(.*)\"\)'
    new_file  = []

    for line in lines:
        line = line.strip()

        res =, line)
        if res:
            new_line = re.sub(pattern, 
                lambda match: '\"' +, + '\"'
            new_line = line
    return new_file

def remove_concat(lines):
    """Replaces the VB concatenation sign "&" with the result string.

        lines (list): List of lines.

        list: List of modified lines.
    pattern = r'\"(.*)\"\&\"(.*)\"'
    new_file  = []

    for line in lines:
        line = line.strip()

        res =, line)
        if res:
            new_line = re.sub(pattern, 
                lambda match: '\"' + + + '\"'
            new_line = line
    return new_file

def remove_trailing_zeros(lines):
    """Removes all trailing NULL bytes from a file.

        lines (list): List of lines.

        list: List of modified lines.
    new_file  = []

    for line in lines:
        if len(line) > 0 and line[0] == '\x00':
    return new_file

def deobfuscate_file(fpath_in, fpath_out):
    """Converts Qbot VBS script into it's deobfuscated form.
    Main changes are:
    - removing stub calculations
    - converting "chr(*)" into their respective characters.
    - converting "replace(*)" into its respective string.
    - converting VB concatenations ("&") into the final string.
    - removing trailing NULL bytes.

        fpath_in (str): Input VBS file path.
        fpath_out (str): Output file path.
        with open(fpath_in, 'r') as f_in:
            in_data =
        return None

    lines = in_data.split('\n')

    lines = remove_additions(lines)
    lines = remove_chr(lines)
    lines = remove_replace(lines)
    for _ in range(100):
        lines = remove_concat(lines)
    lines = remove_trailing_zeros(lines)

    new_file_joined = '\n'.join(lines)

        with open(fpath_out, 'w') as f_out:
        return None

def decrypt_data(enc_str, keys):
    """Decrypts long blob of text data.
    decryption method is looking for patterns of decimal numbers,
    and xor them with the key.
    do that with three different keys.

        enc_str {string} -- encrypted data
        key1 {int} -- first key
        key2 {int} -- second key
        key3 {int} -- third key
    def _decrypt_data_inner(str_param, key_param):
        """Helper method. actual decryption.
        numbers = ""
        ret_decrypted = ""
        f = True
        for i in range(len(str_param)):
            if  '0' <= str_param[i] <= '9':
                numbers = numbers + str_param[i]
                f = True
                if f:
                        enc_ch = int(numbers)
                    dec_ch = enc_ch ^ key_param
                    ret_decrypted = ret_decrypted + chr(dec_ch)
                numbers = ""
                f = False
        return ret_decrypted

    key1 = keys[0]
    key2 = keys[1]
    key3 = keys[2]
    enc_str = _decrypt_data_inner(enc_str, key1)
    enc_str = _decrypt_data_inner(enc_str, key2)
    return _decrypt_data_inner(enc_str, key3)

class qbot_vbs(object):
    """Encapsulates qbot VBS artifacts.
    These artifacts are used for extraction.
    num_urls = 0
    enc_str = None
    key_str = None
    seed = 0
    key_idxs = [None, None, None]

    def __init__(self, data): = data
        self.lines = data.split('\n')

    def _extract_number_urls(self):
        """Extracts number of encrypted urls.
        # sample:
        # number of urls: for i=1 to 6
        pattern = r'[F,f]or i=1 to (\d+)'
        res =, '\n'.join(self.lines))
        self.num_urls = int(

    def _extract_enc_str(self):
        """Extracts the string which has the encrypted data.
        should be the biggest line in the script.
        max_len = 0
        max_idx = -1
        for i, line in enumerate(self.lines):
            if len(line) > max_len and line[0] != '\x00':
                max_len = len(line)
                max_idx = i
        # removing variable name.
        res ='^\w+ = \"(.*)\"$', self.lines[max_idx])
        self.enc_str =

    def _extract_key_str(self):
        """Extracts the string which the key is based upon.
        should be called after `extract_enc_str()`.
        should be the second biggest line after encrypted string.
        second_max_len = 0
        second_max_idx = -1
        max_len = len(self.enc_str)
        for i, line in enumerate(self.lines):
            if len(line) > second_max_len and len(line) < max_len and line[0] != '\x00':
                second_max_len = len(line)
                second_max_idx = i
        # removing variable name.
        res ='^\w+ = \"(.*)\"$', self.lines[second_max_idx])
        self.key_str =

    def _extract_seed_and_key_indexes(self):
        """Helper function for key extraction.

        def _find_variables(var1_name, var2_name):
            """Finds variables values for two vars.
            for example:
            DgZlWOk = 8
            jRryhge = 4
            pattern1 = fr'^{var1_name} = (\d+)$'
            pattern2 = fr'^{var2_name} = (\d+)$'
            var1_value = None
            var2_value = None

            for line in self.lines:
                res =, line)
                if res:
                    var1_value =
                res =, line)
                if res:
                    var2_value =

            return var1_value, var2_value

        def _extract_key_indexes(lines):
            # we have 6 'Mid' encounters:
            # yaGlYs = Mid(xHAaMv, 10, 2)
            # RLquKjB = Asc(Mid(HIbAriX, seSclZ, 1))
            # three times.
            # the second is not interesting for us.
            text = '\n'.join(lines)
            res = re.findall(r'Mid\(\w+\, (\w+)\, \w+\)', text)
            # the key creation order is reversed to their using. (first key3 is set and so on)

            self.key_idxs[0] = int(res[4])
            self.key_idxs[1] = int(res[2])
            self.key_idxs[2] = int(res[0])

        # sample line:
        # For uLRYNs = 0 To 2387414 Step 1
        pattern1 = r' (238\d\d\d\d) '

        # sample line:
        # YLTCm = YLTCm + DgZlWOk - jRryhge
        pattern2 = r'^(\w+) = (\1) \+ (\w+) - (\w+)$'

        for i, line in enumerate(self.lines):
            res =, line)

            if res:
                num = int(


                for inner_line in self.lines[i+1:i+10]:
                    res =, inner_line)
                    if res:
                        first_param =
                        second_param =
                        first_value, second_value = _find_variables(first_param, second_param)

                        num += 1
                        num *= (int(first_value) - int(second_value))
                        self.seed = num

    def extract_keys(self):
        """Main extraction method.
        Extracts keys for the URL decryption.

            list: List of 3 keys.

        seed = self.seed * 999999
        str_seed = str(seed)

        idx = int(str_seed[self.key_idxs[2] - 1:self.key_idxs[2] + 1])
        key3 = ord(self.key_str[idx - 1])

        idx = int(str_seed[self.key_idxs[1] - 1:self.key_idxs[1] + 1])
        key2 = ord(self.key_str[idx - 1])

        idx = int(str_seed[self.key_idxs[0] - 1:self.key_idxs[0] + 1])
        key1 = ord(self.key_str[idx - 1])

        return key1, key2, key3

if __name__ == "__main__":
    if len(sys.argv) != 2:
        print(f"Usage: python {os.path.basename(__file__)} <fpath_in>")

    fname_tmp = 'tmp'

    deobfuscate_file(sys.argv[1], fname_tmp)

    if not os.path.exists(fname_tmp):
        print("Failed de-obfuscation script.")

    with open(fname_tmp) as f:
        data =


    vbs = qbot_vbs(data)
    keys = vbs.extract_keys()
    dec = decrypt_data(vbs.enc_str, keys).strip('\ufeff').split('_______')

    for i in range(vbs.num_urls):
        url = dec[i]
        url = url.split('?')[0].strip()

Appendix C: JavaScript Updater URL Extraction Script

import re
import os

def extract_urls_from_js_updater(js_data):
    """Extracts update URLs out of given Qbot Javascript updater.

        js_data (str or bytes): Javascript code content.

        list: Returns list of extracted URLs or None if falied.

        if isinstance(js_data, bytes):
            js_data = js_data.decode('ascii')
        return None

    arrays = []

    # var WcrApaqyDNEBJYsFkiXPVzHCeKGmnxd = [30,209...19];
    # encrypted urls
    pattern = re.compile(r"^\s*var [a-zA-Z0-9]+\s?=\s?\[(([0-9]+,)+)([0-9]+)\];$")

    for line in js_data.splitlines():
        match = pattern.match(line)
        if match:
            array = +

    if not len(arrays) == 2:
        return None

    suffix = 'datacollectionservice.php3'

    # encrypted text
    base_values = [int(c) for c in arrays[0].split(",")]
    # key
    xor_values = [int(c) for c in arrays[1].split(",")]

    res = ""
    for i in range(len(base_values)):
        res += chr(base_values[i] ^ xor_values[i % len(xor_values)])

    servers = res.split(";")
    urls = ['http://' + server + '/' + suffix for server in servers]
    return urls




  • Check Point Research Publications
  • Global Cyber Attack Reports
  • Threat Research
February 17, 2020

“The Turkish Rat” Evolved Adwind in a Massive Ongoing Phishing Campaign

  • Check Point Research Publications
August 11, 2017

“The Next WannaCry” Vulnerability is Here

  • Check Point Research Publications
January 11, 2018

‘RubyMiner’ Cryptominer Affects 30% of WW Networks