CNCMachineRMS: The Missing Piece in the ROMULUS Campaign

Authors: Sam Decker, Nevan Beal, and Andi Ursry

Key Takeaways

  • Blackpoint’s Adversary Pursuit Group (APG) identified the previously unnamed RAT delivered through ROMULUS as CNCMachineRMS.
  • APG tracked widespread ClickFix activity involving multiple malware delivery chains and variations impersonating legitimate software.
  • The ROMULUS platform distributed both CNCMachineRMS and LegionLoader through separate infection chains.
  • APG recovered the final RAT from a malicious Stardock installer that used DLL sideloading and encrypted payloads.
  • The malware uses a proprietary storage format and custom scripting engine to manage its execution and embedded resources.
  • The malware shared distinctive components with LevelBlue’s earlier CNCMachineRMS research, including family-specific identifiers within the final RAT.
  • CNCMachineRMS supports persistent remote access, surveillance, system reconnaissance, and additional payload execution.
  • ROMULUS uses multiple delivery methods and malware payloads, increasing the risk of persistent access, data theft, and further compromise across business and MSP environments.

Executive Summary

Beginning in late September 2026, Blackpoint’s Adversary Pursuit Group (APG) tracked a high volume of ClickFix activity involving multiple malware delivery chains and installers impersonating legitimate software. On September 28, 2026, researchers at Huntress published an investigation into the same activity, documenting the abuse of ChatGPT Custom GPTs to distribute malicious installers. Their research detailed a complex infection chain involving DLL sideloading, concealed shellcode, and encrypted storage, ultimately delivering an unidentified remote access trojan (RAT). APG observed several variations of this activity, including delivery chains abusing legitimate applications from Canon and Stardock.

On October 15, 2026, researchers at Forescout’s Vedere Labs expanded on these findings, connecting infrastructure associated with the Stardock delivery chain to a previously undocumented malware delivery platform named ROMULUS. Their investigation identified multiple infection chains supported by the platform, including the previously documented Stardock installer and a separate delivery route involving LegionLoader. APG also observed both delivery routes during its investigation, further connecting the activity it had been tracking to the broader ROMULUS campaign.

While the research published by Huntress and Forescout provided extensive insight into the campaign’s delivery infrastructure, infection chains, and the capabilities of the final RAT, the implant had yet to be attributed to a known malware family. Throughout its investigation, APG independently analyzed multiple payloads distributed through the campaign, including variations of the infection chain originally documented by Huntress.

Further analysis revealed strong similarities to CNCMachineRMS, a RAT documented by LevelBlue SpiderLabs in August 2026 as the final payload of a separate BabaDeda infection chain. The connection was supported by the proprietary storage format, scripting engine, and execution routines, along with identifiers specific to CNCMachineRMS recovered directly from the final implant. Based on these findings, APG identified the previously unnamed RAT delivered through ROMULUS as CNCMachineRMS, alongside the platform’s separate LegionLoader delivery chain.

Assessment

Blackpoint’s APG assesses with high confidence that the unnamed RAT delivered through the ROMULUS Stardock chain is CNCMachineRMS. A matching payload SHA-256, family-specific strings, and matching storage structures support this assessment. The findings support shared tooling across the BabaDeda and ROMULUS ecosystems; however, they do not establish a shared or connected operator.

ROMULUS likely operates as a delivery service, distributing CNCMachineRMS and LegionLoader through separate chains while retaining ClickFix and signed binary sideloading techniques. Defenders should prioritize recurring behaviors, including PowerShell to MSI execution, dependency sideloading from user-writable paths, combined Run Key and scheduled task persistence, and non-browser DNS over HTTPS traffic. CNCMachineRMS supports surveillance and additional payload execution, but no post-exploitation activity has been observed.

Blackpoint’s APG assesses with high confidence that service delivery platforms, like ROMULUS, will continue to be developed and offered across cybercriminal ecosystem as the ClickFix-style techniques continue to evolve and prove an effective initial access vector. It is likely that over the next 12 months, tools like ROMULUS will continue to be improved and deliver additional variants as they are increasingly adopted by threat actors.

Connecting the Research

The investigations published by Huntress, Forescout, and LevelBlue SpiderLabs provide extensive technical coverage of the individual infection chains, delivery infrastructure, and malware components discussed in this article. Rather than repeating that analysis, APG’s research focuses on the similarities that connect these findings and establish CNCMachineRMS as one of the payloads distributed through ROMULUS.

Throughout the campaign, APG encountered multiple variations of the delivery chain, including installers impersonating different legitimate applications. The Stardock variant examined below represents just one of these variations and serves as the primary example for demonstrating the connection to CNCMachineRMS. While the delivery components may differ, the underlying implementation provides a common point of comparison across the activity.

Technical Analysis

From ClickFix to a Trusted Host Process

Among the payloads analyzed by APG was IconEdit2Turb.msi, a malicious installer distributed through the ClickFix activity associated with ROMULUS. The delivery chain used an obfuscated PowerShell stager to retrieve the installer, remove its Mark of the Web (MotW), and execute it silently through msiexec.exe. The download infrastructure also used decimal IP notation, a technique APG has observed across other ClickFix campaigns.

The MSI masqueraded as Stardock Smart DeElevation Tool, installing its components under %LOCALAPPDATA%\Programs\Stardock Smart DeElevation Tool\. Analysis of the installer revealed that it set ARPSYSTEMCOMPONENT=1, hiding the application from conventional installed program listings. Once installation completed, a WixShellExec custom action launched DeElevate64.exe, a legitimate, digitally signed Stardock executable (Figure 1).

Figure 1: MSI custom action launching DeElevate64.exe.
Figure 1: MSI custom action launching DeElevate64.exe.

Rather than modifying the signed executable, the attacker altered its companion library, DeElevator64.dll, to introduce an additional dependency on I++u.dll, a malicious DLL included in the installation directory. When DeElevate64.exe starts, Windows loads the modified companion DLL and its additional dependency, allowing the malicious code to execute within the legitimate Stardock process (Figure 2).

Figure 2: Modified DeElevator64.dll importing the malicious I++u.dll.
Figure 2: Modified DeElevator64.dll importing the malicious I++u.dll.

A Loader Hidden Inside a NuGet Package

Once I++u.dll is loaded, it creates a thread that reads an encrypted payload from Build.dat. Although the file appears to be a legitimate Microsoft System.Net.Http NuGet package, APG discovered that the malicious payload was concealed within the compressed data of an existing package entry, rather than stored as a separate file. This allowed the shellcode to remain hidden within what otherwise resembled an ordinary software package (Figure 3).

Figure 3: Simplified pseudocode showing I++u.dll reading the concealed payload from Build.dat.
Figure 3: Simplified pseudocode showing I++u.dll reading the concealed payload from Build.dat.

Further analysis of I++u.dll revealed a position dependent XOR routine used to decrypt the embedded payload. APG reconstructed this routine and successfully recovered approximately 341 KB of x64 shellcode (Figure 4). The malware then uses senddmp.resources.dll to allocate executable memory through VirtualAlloc. The loader passes the shellcode’s address to EnumSystemCodePagesW, using the Windows API’s callback mechanism to execute the decrypted code in memory.

Figure 4: Simplified pseudocode of the loader's position dependent XOR decryption routine.
Figure 4: Simplified pseudocode of the loader’s position dependent XOR decryption routine.

The recovered shellcode handles the next stage of execution, including accessing execute_engine_disconnect.raw, a proprietary obfuscated Storage file containing the resources used to load the final implant. Analysis of this storage format would ultimately provide some of the strongest evidence connecting the payload to CNCMachineRMS.

Opening the Custom Storage File

The decrypted loader accesses execute_engine_disconnect.raw, a proprietary storage file containing the resources needed to execute the final implant. Rather than using a conventional archive format, the malware organizes its contents into a hierarchy of directories, files, and properties. The loader applies a series of arithmetic operations to decode the archive’s metadata, while individual file contents are obfuscated through a separate bytewise XOR routine (Figure 5).

Figure 5: Simplified pseudocode showing arithmetic decoding of the proprietary Storage metadata.
Figure 5: Simplified pseudocode showing arithmetic decoding of the proprietary Storage metadata.

APG reverse engineered the loader’s storage routines and reconstructed the archive parser, allowing us to extract its contents without executing the malware. The decoded archive contained 1,128 entries, consisting of 315 directories and 813 files. The recovered record boundaries and file sizes accounted for the entire archive, confirming that the parser was correctly interpreting its internal structure.

Among the extracted resources were two components of particular interest. The first, identified as class 16001, contained a 3,904-byte controller script responsible for managing the malware’s execution and persistence. APG successfully decoded this script, revealing several tasks that maintain persistence and load the final implant. The second resource, class 16100, contained approximately 1.58 MB of native x64 shellcode representing the final RAT payload (Figure 6).

Figure 6: Recovered controller script loading the embedded RAT.
Figure 6: Recovered controller script loading the embedded RAT.

The recovered RAT’s SHA-256 also matched the final implant documented by Huntress, confirming that both investigations recovered the same payload.

What the Implant Provides

The final payload is a fully featured remote access trojan capable of maintaining persistent access to compromised systems. Analysis of the same implant by Huntress documented capabilities including remote desktop control, screen streaming, microphone and camera capture, file management, and system reconnaissance. The RAT also supports executing additional payloads, allowing operators to deploy further malware or tools following the initial infection.

Much of the malware’s functionality is controlled through a custom scripting engine, which manages persistence, communications, resource loading, and code execution. Rather than relying on a fixed execution sequence, the framework uses scripts to invoke native functions and access resources stored within its proprietary archive.

APG’s analysis of the recovered controller script also revealed how the malware maintains persistence. Separate tasks periodically check and recreate a Registry Run entry and scheduled task at intervals of 150 and 875 seconds, respectively. A shutdown monitor provides an additional mechanism for restoring Registry persistence. Both autorun mechanisms use the name Stardock DeElevation Tool.

For command and control, the RAT supports DNS over HTTPS (DoH) to resolve its active Gate, allowing it to discover its command and control server through public DNS resolvers rather than relying solely on the system’s configured DNS servers.

Connecting the Dots to CNCMachineRMS

With the final payload recovered, APG compared its underlying implementation against research published by LevelBlue SpiderLabs in August 2026, which documented CNCMachineRMS following an investigation into a separate BabaDeda infection chain. Although LevelBlue’s investigation involved DLL sideloading through the legitimate IBM SPSS executable WinWrapIDE.exe, the malware’s internal components shared several distinctive characteristics with those recovered by APG.

The strongest similarities were found in the proprietary Storage format used by both samples. The archives shared the same header and record structures, arithmetic operations for decoding metadata, and separate transformations for extracting their embedded resources. APG’s analysis also identified the same resource classes documented by LevelBlue, including a class 16001 scripting resource containing exactly 3,904 bytes in both samples. In each case, class 16100 contained the final RAT payload, although the extracted shellcode differed in size between the two variants.

ComponentStardock SampleLevelBlue Sample
DeliveryStardock DeElevate64.exeIBM SPSS WinWrapIDE.exe
Storage format0x78-byte header, 0x58-byte recordsSame structure
Metadata protectionArithmetic field decoding and bytewise resource transformationSame approach
Storage record count1,1281,128
Script resourceClass 16001, 3,904 bytesClass 16001, 3,904 bytes
RAT resourceClass 16100Class 16100

Beyond the storage format, LevelBlue also documented a custom scripting engine responsible for managing persistence, communications, resource loading, and execution. This helps explain the role of the scripting resource recovered by APG and further supports the connection between the two implementations.

Further analysis of the final implant revealed a more direct connection to CNCMachineRMS. The RAT constructs several strings at runtime using an embedded decoding routine, including CNCMachineRMS\tasks and task_payload.bin. These strings match the task staging structure documented by LevelBlue SpiderLabs in its August research, providing additional evidence of the shared codebase within the final RAT itself (Figure 7).

Figure 7: Reconstructed string decoding routine revealing the CNCMachineRMS task directory.
Figure 7: Reconstructed string decoding routine revealing the CNCMachineRMS task directory.

These similarities extend well beyond the common DLL sideloading and shellcode execution techniques used by both infection chains. The matching proprietary Storage format, resource organization, and execution architecture, combined with the CNCMachineRMS identifiers recovered directly from the final implant, allowed APG to attribute the previously unnamed RAT delivered through ROMULUS to CNCMachineRMS, connecting the campaign to the malware family previously documented by LevelBlue.

Conclusion

APG’s investigation identified CNCMachineRMS as the previously unnamed RAT delivered through the ROMULUS campaign, alongside the separate LegionLoader delivery chain. By examining the underlying malware implementation and identifying similarities with the research published by LevelBlue SpiderLabs in August 2026, APG was able to connect the activity documented by Huntress and Forescout to an established malware family. These findings help complete the picture of the malware being distributed through ROMULUS and establish a connection between previously separate investigations.

Throughout the campaign, APG observed multiple variations of the infection chain involving different installers and legitimate applications. While this analysis focused on the Stardock variant to demonstrate the connection to CNCMachineRMS, the broader investigation extended beyond this single delivery method. As attackers continue to change the applications and loaders used to distribute their payloads, the underlying malware architecture provides a more consistent way to identify related activity and understand the threats being delivered.

Recommendations

  • Strengthen application controls. Restrict unauthorized software installations and execution from untrusted locations. Use application allowlisting to reduce exposure to malicious installers and DLL sideloading.
  • Limit unnecessary privileges. Apply least privilege principles and restrict local administrator access to reduce opportunities for unauthorized system changes.
  • Harden endpoint security. Ensure endpoint protection, behavioral monitoring, and tamper protection are enabled and consistently enforced across managed systems.
  • Improve user awareness. Educate users about ClickFix attacks, particularly fake verification prompts that instruct them to execute commands through PowerShell or Windows Run.
  • Monitor suspicious execution chains. Investigate PowerShell downloading scripts or installers, particularly when followed by silent MSI execution or unusual child processes.
  • Detect DLL sideloading. Monitor legitimate signed applications executing from unexpected directories, especially when loading modified or unsigned DLLs.
  • Review persistence mechanisms. Hunt for suspicious Registry Run entries and scheduled tasks, particularly when both are used to maintain access or automatically restore persistence.
  • Restrict unauthorized network communications. Implement DNS filtering and outbound traffic controls, and investigate suspicious DNS over HTTPS activity associated with unexpected processes.

How Blackpoint Reduces Exposure

Blackpoint’s Security Operations Center (SOC) continuously monitors for suspicious activity across managed environments, including the execution patterns, unauthorized software installations, and persistence mechanisms associated with ClickFix campaigns. By correlating endpoint activity across different stages of an infection, Blackpoint analysts can identify malicious behavior even when attackers change their delivery methods or payloads.

Blackpoint’s Adversary Pursuit Group (APG) supports these efforts through malware analysis, threat research, and targeted investigations. By reverse engineering the payloads distributed through ROMULUS and identifying the previously unnamed RAT as CNCMachineRMS, APG provides additional insight into the malware’s capabilities and underlying architecture. These findings help inform threat hunting, detection engineering, and incident response efforts across Blackpoint’s managed environments.

Indicators of Compromise

File Hashes

ArtifactSHA-256Context
ISOSimple.msi6761aad48a3f987238994d92bca97e4b8550e0150607bd67b47b1b6366a371fcMalicious installer
IconEdit2Turb.msi91a22cf3154944897cbcaffc7d20d4596e972d280ee134197d41d8d8eefb0fe2Malicious installer
FilesInspectorX.msi9ac13a8ef7bab76c4d074385fd783e8c9542a8db7a166c3d3921a48add092455Malicious installer
ClipboardFusion.msi917f2bc33faba15236be1bf6c5a121dbca3fdf38b56c03ad08e6c46b464509feMalicious installer
AllMyBasic.msif4d7c5321bb0d1bda439304a753bcf9b467b98ca53c7e9493b9c2c3451092e4aMalicious installer
DeElevator64.dll0457414c4504b70115798eee9c8384a8bf9e793461ffb2e0661a6dcc6ed4809fModified sideloaded DLL
I++u.dll3cd1484cc5bf10e22d79784beba58a75d7b126c46efb5e1a85d6aab884447621Malicious loader dependency
senddmp.resources.dllb59a21a8d4c0c11a9ecbbdbe2a2938499c99c5ebb210c9c6db9683c7ad19f4c4Executable-memory helper
res.dlla27cafb84876a3374ef4aba1062345def6cd0b029eaf8e0dc44fa24fc943f3c6Additional loader component
Build.datd3fda1e6cf886fd530bf62e2d16c39ad23ed12a2c933fc211f1f4876ceaa033eConcealed shellcode carrier

Host and Persistence Indicators

IndicatorContext
%LOCALAPPDATA%\Programs\Stardock Smart DeElevation Tool\Malicious installation directory
HKCU\Software\Microsoft\Windows\CurrentVersion\RunRegistry persistence using the value Stardock DeElevation Tool
Stardock DeElevation ToolScheduled task used for persistence
CNCMachineRMS\tasksFamily specific task directory constructed by the RAT
task_payload.binTask payload filename identified in the RAT

Network Indicators

IndicatorContext
167.76.44[.]160ROMULUS C2 server; also served ClipboardFusion.msi
167.148.201[.]224ROMULUS C2 server; also served ClipboardFusion.msi
93.152.213[.]151ROMULUS C2 server
96.62.224[.]81ROMULUS C2 server
151.240.151[.]223ROMULUS C2 server
86.107.124[.]7ROMULUS C2 server
86.109.75[.]7ROMULUS C2 server
23.94.252[.]153ROMULUS C2 server
2.27.248[.]127ROMULUS C2 server
cloubflare[.]topROMULUS C2 server
cloubflare[.]netROMULUS C2 server
hxxp://167.76.44[.]160/app/5dbd80d34839/ClipboardFusion.msiClipboardFusion.msi delivery URL
hxxp://167.148.201[.]224/app/5dbd80d34839/ClipboardFusion.msiClipboardFusion.msi delivery URL

References

Subscribe to receive updates from Blackpoint’s Adversary Pursuit Group

DATE PUBLISHEDNovember 18, 2026
AUTHORSam Decker, Nevan Beal, and Andi Ursry

The Cost of Waiting

The Cost of Waiting: What Identity Attacks Cost the MSP ebook What Identity Attacks Cost the MSP Stolen credentials don't trip alarms. Once an attacker is authenticated, they look like anyone else, and most tools aren't watching what happens next. Learn what closing the identity response gap looks like: how identity attacks happen, where response slows down, and the questions to ask any ITDR vendor.
*Identity-based techniques play a role in nearly 90% of incident investigations

GET YOUR COPY