Contributor: Jason Zhang
Emotet attacks leveraging malicious macros embedded in Excel files continue, with new variants and novel tactics, techniques, and procedures (TTPs). Following our recent report, we observed new waves of Emotet campaigns abusing legitimate Windows features, such as batch scripts and the mshta utility, combined with PowerShell, to deliver Emotet payloads.
In this follow-up blog post, we first provide an overview of the delivery processes of Emotet payloads in typical attacks. Then, we examine the recent variants and reveal how techniques evolved in these attacks.
Emotet payload delivery chain
The Emotet infection chain typically starts with a spam email containing a malicious document in the attachment (see Figure 1). The attachment can be either a Word document or an Excel file with embedded VBA or Excel 4.0 (XL4) macros. To entice the user to enable macro execution in Microsoft Word or Excel, the file displays social engineering content when opened. Once macro execution has been enabled, the embedded macro is executed, leading to the delivery process of an Emotet payload.
As highlighted in Figure 1, there are typically two ways to deliver an Emotet payload:
- Executing macros to download the payload from a remote server directly: this can be achieved by invoking the WinAPI URLDownloadToFile via the CALL function, as we reported earlier.
- Invoking PowerShell scripts execution to download the payload: the PowerShell scripts can either be embedded in macros or downloaded from a remote server. Attackers can leverage various adversary techniques, such as multiple layers of obfuscation, to hide the PowerShell scripts. This can increase the infection rate by evading anti-virus (AV) scanners, particularly signature-based detection engines. As a result, using PowerShell scripts to download Emotet payloads has become a favorite choice in these attacks. The attacks discussed in this report are examples of this scenario.
Figure 2 shows the detection timeline of a few recent Emotet waves attacking some of our customers. The first attack started on January 11, followed by two major waves afterwards.
Based on our investigation, we can classify these attacks into three clusters (marked as A, B, and C in Figure 2):
- Cluster A: Emotet payload via XL4 macro directly
- Cluster B: Emotet payload via XL4 macro with PowerShell
- Cluster C: Emotet payload via VBA macro with PowerShell
The Cluster A attack leveraged the XL4 macro to download the Emotet payload directly from one of several possible remote servers, as has already been discussed in our earlier report. In the following sections, we will focus our analysis on clusters B and C. More specifically, we’ll investigate the techniques used in clusters B and C to understand how the Emotet payload is delivered differently, though the opening pages from the samples of both clusters appear very similar (see Figure 3).
Cluster B: Emotet payload via XL4 macro with PowerShell
To investigate Cluster B, we analyzed one of the samples from this campaign (see Table 1).
Figure 4 shows the extracted XL4 macro from the sample. The macro is stored in a macro worksheet called KEY. The macro is relatively simple, with less obfuscation compared to samples in the Cluster A wave (see report).
When macro execution is enabled, the macro script runs cmd.exe via the EXEC function to spawn a process to execute mshta (the de-obfuscated version of m^sh^t^a from the SET.NAME command line shown in the figure). Once mshta is launched, it executes a remote file. The remote file URL is obfuscated with a hexadecimal-encoded server name:
De-obfuscating it reveals the actual URL:
Next, we discuss mshta and investigate the content stored in the file above.
mshta.exe is a 20-year-old Windows-native utility that executes Microsoft HTML Application (HTA) files. Windows-legitimate tools like mshta and PowerShell, dubbed “LOLBINs” (living-off-the-land binaries), have become increasingly popular among threat actors. This is because these utilities and scripts are signed by Microsoft and trusted by the Windows OS, allowing attackers to bypass detection by proxying execution of the malware. MITRE reports T1218 and T1216 provide more information on signed binary proxy execution and signed script proxy execution, respectively.
Attackers typically abuse mshta to execute malicious VBScript and Jscript programs either through inline commands (e.g., Kovter Trojan), or through an HTA file. The attack we investigate herein is the latter case. The next stage payload is stored in the remote HTA file and delivered via mshta execution. A copy of the HTA file is also downloaded to the C:\Users\ directory during the execution process, as confirmed by analyzing the Excel sample with VMware NSX Advanced Threat Analyzer (NSX ATA, see Figure 5).
At a first glance, the HTA file (sha1: 2615f7aa2141cc1cb5d0c687bc3396981c2c68dc) doesn’t appear to contain anything, and it is empty when opening it in a common editor. Using whitespace obfuscation is a typical example of a steganography technique (see MITRE report T1027.003) to hide code or make analysis harder. One can use tools such as js-beautify to remove the empty lines and prettify the script inside. Figure 6 shows the first and last parts of the prettified JScript contained in the HTA file.
The JScript is highly obfuscated. At the end of the script, the unescape() and eval() functions (as highlighted in the figure above) are called to decode and execute the obfuscated script. Executing the Jscript sample within VMware NSX ATA shows that a new process is spawned to invoke PowerShell execution.
Our analysis reveals that the remaining process to deliver the final Emotet payload involves two stages:
- The PowerShell script contained in the HTA file downloads another PowerShell payload from a remote URL.
- The downloaded PowerShell script from the previous step downloads the Emotet DLL payload.
PowerShell: first stage
Figure 7 shows the PowerShell payload contained in the HTA file.
After removing the obfuscating strings, the purpose of the script becomes more noticeable – when the PowerShell script is executed, it attempts to download another payload using the .NET WebClient.DownloadString method, as highlighted in Figure 8. The IEX command (shown at the end of the figure) is an alias for the Invoke-Expression cmdlet that evaluates and runs the string specified by the $JI variable. One can ignore the backticks, as they are used just to obfuscate the command.
PowerShell: second stage
While the payload (sha1: dcc120c943f78a76ada9fc47ebfdcecd683cf3e4) downloaded from the previous stage has an image file extension .PNG, in fact it is another PowerShell script but without obfuscation (see Figure 9). This time it calls the .NET WebClient.DownloadFile method to download the Emotet DLL payload from one of 10 hosts (as compared to 3 hosts seen in samples from Cluster A) and save it to C:\Users\Public\Documents\ssd.dll (sha1: e597f6439a01aad82e153e0de647f54ad82b58d3).
At the end, the process pauses for 4 seconds by running Sleep -s 4 (see Figure 9). This is to make sure the payload is properly saved before calling cmd.exe to launch rundll.32.exe and execute the Emotet DLL payload.
Cluster C: Emotet payload via VBA macro with PowerShell
Similarly, we analyzed one of the samples from the Cluster C campaign (see Table 2).
Like most Emotet attacks seen in the past, the file contains highly obfuscated VBA macros. If macro execution is enabled, the embedded VBA scripts are executed. Figure 10 shows a snippet of the macros.
To reveal the actual malicious behavior, one can use open-source emulation tools such as ViperMonkey to analyse and de-obfuscate the VBA macros. The emulation results show that the macros will drop and execute a Windows batch script during execution. The batch file is saved to the hidden folder ProgramData on the system drive. This is also confirmed by analyzing the Excel sample with VMware NSX ATA (see Figure 11).
Figure 12 shows the highly obfuscated content of the dropped batch file (sha1: bf1171da149442a511a946ded726b94c1ad7cf85).
It seems that AV engines are not effective at catching fresh malicious batch scripts. In this example, only 2 of 58 AV engines detected the batch file on VirusTotal (see Figure 13, the results were checked five days after they were first submitted to VirusTotal).
After de-obfuscating the batch script with batch deobfuscator, it reveals the PowerShell payload (see Figure 14).
As highlighted in Figure 14, the batch script runs PowerShell in the background with the command start/B. The PowerShell command accepts a base64-encoded string version of a command by using the -enc (or EncodedCommand) parameter. After decoding the base64-encoded string, the de-obfuscated PowerShell payload is shown in Figure 15.
The PowerShell script attempts to download the Emotet DLL payload from one of 12 hosts and save it to the C:\ProgramData\ folder with a randomly generated number as file name, e.g., c:\ProgramData\759027055.dll (sha1: 606e6e0f3b476b2dae26c7f3c95e392a911d04c5). Once the Emotet payload is downloaded successfully, the script calls cmd.exe to launch rundll.32.exe and execute the payload, as highlighted in Figure 15.
VMware NSX detection with MITRE ATT&CK mapping
VMware NSX customers are well-protected against Emotet attacks from both clusters. Figure 16 and Figure 17 show the analysis overview from VMware NSX ATA when executing the initial Excel samples listed in Table 1 and Table 2, respectively. As shown in both figures, NSX ATA successfully detected both samples as Emotet. While NSX ATA identified a few common high-risk characteristics from both clusters, such as the spawning of shell command and PowerShell, the observation of command & control traffic, and the execution of a dropped a file, there are also some techniques that distinguish both clusters. For example, the sample from Cluster B exhibits the behaviors of running XL4 macro with suspicious external commands and spawning an mshta process (see Figure 16), while the Cluster C sample is capable of dropping and executing a script file (see Figure 17).
The analysis overview also contains MITRE ATT&CK tactic and technique mapping for some of the key malicious behaviors observed during the attack execution (as highlighted in both Figure 16 and Figure 17). The typical ATT&CK tactics used in this attack include TA0002: Execution, TA0005: Defense Evasion, and TA0011: Command and Control. A detailed MITRE ATT&CK tactic and technique mapping for Emotet can be found in the MITRE report.
The increasing challenges the security community faces today not only come from new threats, but also from old malware armed with new TTPs. The recent Emotet attacks discussed in this blog post and our previous report are notable examples of these long-term threats that leverage legitimate Windows features (such as XL4 macros and batch scripts) and LOLBINs (like mshta and PowerShell utilities). These attacks impose great challenges on traditional AV solutions, in particular signature-based detection systems. In contrast, behavior-based detection systems such as VMware’s AI-driven NSX ATA are very effective in successfully identifying the techniques leveraged by attackers.
Indicators of compromise identified from this report can be found on VMware TAU’s GitHub IoCs repository.