1. Description
Operation FlashHole is a targeted cyber-espionage campaign in which the Lazarus Group reportedly exploited legitimate South Korean software and deployed a sophisticated, multi-stage malware framework.
According to Kaspersky, the attackers abused CrossWeb EX, a financial-security application used to support access to online banking and government websites in South Korea. Although the precise mechanism that initiated the infection remained unclear, researchers reconstructed an execution chain involving malicious code injected into a legitimate Windows process and multiple subsequent malware stages.
The campaign combines established Lazarus-associated tools with newly identified malware families. TechHub acts as an initial downloader and uses GitHub for command-and-control communications. Cubridge coordinates additional payloads and modules, while TouchHub and TouchShift help establish persistence. ClientHub and LPEClient provide further capabilities for host profiling, payload delivery, and concealed communication.
Kaspersky identified at least three affected organizations in South Korea and assessed that the true number of impacted organizations may be larger. The researchers characterize Operation FlashHole as an espionage-focused operation, distinguishing it from other contemporaneous activity that involved data exfiltration and apparent monetization.
2. Technical Description
Threat actor: Lazarus Group, as assessed by Kaspersky
Campaign: Operation FlashHole
Broader campaign cluster: Cascade Hole
Threat category: Advanced persistent threat (APT), targeted cyber-espionage, malware deployment
Primary operating environment: Microsoft Windows
Initial access and delivery mechanisms reported:
Targeted phishing emails containing links to fabricated portfolio or résumé websites.
A watering-hole-style approach involving malicious websites.
Abuse of legitimate South Korean financial-security software.
Malware families identified:
TechHub
Cubridge
TouchHub
TouchShift
ClientHub
LPEClient
Supporting infrastructure and techniques:
GitHub-based command-and-control communication.
A Dropbox API-based communication variant.
DLL side-loading and process injection.
Reflective loading and memory-resident payload execution.
Windows service installation for persistence.
Encrypted payload exchange and modular malware delivery.
Reported Kaspersky detection verdicts:
MEM:Trojan.Win64.Shellcode.genTrojan.Win64.DLLhijack.*Trojan.Win64.Lazarus.*
3. Attack Overview
3.1 Targeted delivery
The operators reportedly used spear-phishing emails containing links to fabricated websites tailored to intended victims. The sites included résumé and interior-design portfolio content that appeared relevant to the targeted organizations.
3.2 Abuse of trusted software
The infection chain involved CrossWeb EX, a legitimate financial-security application used in South Korea. Kaspersky observed suspicious execution originating from the software environment but could not definitively determine the exact mechanism that initiated the infection.
3.3 In-memory execution
The initial stages involved shellcode and malicious payloads executing in memory. This reduced the amount of malicious code left on disk during the early compromise stages and complicated forensic analysis.
3.4 Persistence and malware deployment
The attackers deployed TouchHub and TouchShift through DLL side-loading and Windows service mechanisms. Subsequent stages introduced ClientHub and LPEClient, enabling additional host profiling and payload delivery.
3.5 Modular espionage capabilities
Cubridge served as a central orchestrator, supporting additional plugins, modules, and tools. The recovered components included capabilities for collecting files and process information, interacting with Windows systems, and extracting browser data.
3.6 Command-and-control communications
The malware used GitHub repositories, compromised websites, and—in one observed ClientHub variant—the Dropbox API to exchange information and retrieve commands or payloads.
4. Technical Analysis
4.1 Initial infection vector
Kaspersky identified suspicious websites hosted on GitHub Pages and associated them with the operation. The sites contained links to external JavaScript resources:
hxxps://javadevhouse.co[.]kr/ajax/libs/jquery/1.7.3/jquery.min.jshxxps://pythonapischool[.]com/api/detach
One website presented a résumé for a supposed senior network-appliance developer, while another featured an interior-design portfolio.
Researchers observed that the résumé photograph lacked EXIF metadata and displayed unusual visual characteristics consistent with possible AI generation. The site's content also contained seemingly artificial details. Kaspersky assessed that AI may have been used to create parts of the lure material before it was tailored to specific targets.
The researchers assessed with medium-to-high confidence that victims received phishing emails linking to the malicious sites. However, they could not precisely determine the event that initiated every infection.
4.2 Abuse of CrossWeb EX
CrossWeb EX is a financial-security application used in South Korea to support secure communication between browsers and web servers, including through public key infrastructure.
Kaspersky observed a suspicious execution chain originating from a legitimate software component associated with CrossWeb EX. The parent process identified in the analysis was:
INISAFECrossWebEXSvc.exe
The observed installation bundle corresponded to INISAFE CrossWeb EX V3 version 3.4.1.2.
The attack subsequently involved the legitimate Windows executable fsquirt.exe, which was executed in the CrossWeb EX process context and injected with shellcode containing TechHub malware.
The exact mechanism by which CrossWeb EX was abused remained unclear. Consequently, the observed chain should not be interpreted as proof that every installation of the legitimate application is compromised.
4.3 TechHub: Initial downloader
TechHub was a central component in the initial malware chain. Unlike earlier versions described in prior research, the variant examined by Kaspersky communicated with a private GitHub repository through the GitHub API.
The malware used a Personal Access Token (PAT), either hard-coded or obtained from its configuration, to interact with the repository.
Its functions included:
Generating or retrieving an identifier for the compromised host.
Establishing a communication handshake with the repository.
Exchanging encrypted information and payloads.
Gathering system information.
Receiving commands and deploying subsequent payloads.
Kaspersky documented three variants: an in-memory version, a DLL side-loading version, and a DLL side-loading version using a registered Windows service.
The GitHub-based approach allowed the malware to use repository issues and files for communication. Its protocol included encrypted key exchange and encrypted file transfers.
4.4 Cubridge: Malware framework orchestrator
Cubridge acted as the central orchestrator of the malware framework. It supported multiple payload types, including plugins, tools, and modules.
The malware could load additional components using techniques such as reflective DLL loading, DLL injection, and section-mapping injection.
Its documented capabilities included:
Gathering device and process information.
Reading, creating, copying, moving, and deleting files.
Executing commands and interacting with Windows processes.
Loading and unloading additional modules.
Collecting browser information.
Capturing screenshots.
Collecting files and directory metadata.
Retrieving and updating configuration information.
Cubridge also generated changing command-and-control URLs using predefined word lists and randomly generated numeric path segments. This made its network requests less uniform and complicated straightforward indicator matching.
Kaspersky linked the examined Cubridge configuration to the following infrastructure:
hxxps://cubedesign[.]co[.]kr/news
The malware's modular architecture allowed the operator to introduce additional functionality without requiring a single monolithic payload.
4.5 TouchHub and TouchShift: Persistence
TouchHub was a previously undocumented component identified by Kaspersky. It installed a legitimate Windows executable under a different filename and placed the malicious TouchShift payload alongside it.
The observed service-related executable path was:
C:\Windows\System32\RasConv\ImeBroker.exe
The side-loaded payload was placed at:
C:\Windows\System32\RasConv\mscoree.dll
The configuration file was:
C:\Windows\System32\RasConv\ActivationStore.dat
TouchHub registered a Windows service and launched it, allowing TouchShift to execute with SYSTEM privileges in the documented chain.
The malware also synchronized file timestamps with those of a legitimate system file, a behavior that could hinder straightforward timeline analysis.
TouchShift decrypted an embedded payload and executed the next TechHub stage. This helped the attackers establish persistence and continue the infection chain after the initial compromise.
4.6 ClientHub: Flexible communication and payload delivery
ClientHub was another newly identified component. Kaspersky observed at least two variants:
A version communicating with websites over HTTP or HTTPS.
A version communicating through the Dropbox API using credentials.
ClientHub supported a modular command mechanism and could load additional DLLs and plugins into memory. Its communication protocol used custom message fields and encrypted data.
Researchers identified plugins with capabilities including:
File collection and metadata gathering.
File archiving.
File cleaning and deletion.
Monitoring drive and remote-session changes.
ClientHub also supported communication with LPEClient, providing a way to relay information and payloads between the malware and command-and-control infrastructure.
4.7 LPEClient: Host profiling and payload delivery
LPEClient was previously associated with Lazarus activity and was also observed in Operation SyncHole.
The updated version described by Kaspersky collected system information, including details about the operating system, hardware, user environment, security products, and Windows security mitigation status.
It interacted with ClientHub through a shared buffer rather than relying on an embedded command-and-control address in the same manner as earlier versions.
Kaspersky also documented changes intended to conceal execution and facilitate payload loading. The malware used multiple cryptographic mechanisms, including GOST, AES, and Threefish, and incorporated ML-KEM-1024 public keys during the payload preparation process.
Researchers ultimately identified Cubridge as the final payload loaded through this sequence.
These findings demonstrate the operation's emphasis on layered execution, host profiling, and concealment. They do not mean every observed sample necessarily executed every capability described in the report.
4.8 Infrastructure and attribution
Kaspersky identified at least three primary GitHub repositories and one fallback repository used for command-and-control operations. The researchers found evidence that GitHub had been used for this activity since at least July 2025.
Other observed infrastructure included compromised South Korean websites and a Dropbox API-based channel used by one ClientHub sample.
Kaspersky assessed with medium-to-high confidence that Operation FlashHole and Operation SyncHole involved the same actor, citing similarities in their infection vectors, targets, malware components, and operational improvements.
The researchers attributed Operation FlashHole to Lazarus and classified it within the broader Cascade Hole campaign. The operation was assessed as focused on espionage, while other concurrent activity involving the COPPERHEDGE backdoor showed indications of data exfiltration and monetization.
5. Indicators of Compromise (IOCs)
The following indicators were published in Kaspersky's Operation FlashHole report. Defenders should preserve the reported defanged formats when documenting these indicators and normalize them carefully when creating detection rules.
5.1 File hashes
The following 32-character hexadecimal values are reproduced as reported in the source's file-hash section.
| Malware/component | File name | Reported hash |
|---|---|---|
| TouchShift | mscoree.dll | ec84d6cb2ae7974767aedac246b50a3e |
| TouchShift | mscoree.dll | a28cba214058fcae2883d8cffc883abd |
| TouchShift | mscoree.dll | c105d6b16f544582772de0159f24c00a |
| TouchShift | mscoree.dll | 2f8475a90a25624ed02f1d18d4e6d65d |
| TouchShift | mscoree.dll | 676fabe30a28b53925fba98e34b35986 |
| TouchShift | mscoree.dll | b4a7eff35f048f06fb0a2d7631f23a7d |
| ClientHub Loader | liblzma.dll | ff07641dc56914dd90ee0bd7404d7ad7 |
Hash format note: the values are 32 hexadecimal characters, consistent with MD5. Verify the hash algorithm against the linked source records before importing them into systems that require an explicitly typed hash.
5.2 File paths
The following paths were listed in Kaspersky's report:
C:\ProgramData\Dell\Dell1.tmp
C:\ProgramData\Dell\Dell3.tmp
C:\ProgramData\Package.txt
C:\ProgramData\RasConv\mscoree.dll
C:\ProgramData\Apple\mscoree.dll
C:\ProgramData\ssh\mscoree.dll
C:\ProgramData\qorestor\mscoree.dll
C:\ProgramData\Dell\mscoree.dll
C:\Windows\System32\RasConv\mscoree.dll
C:\Windows\System32\RasConv\ActivationStore.dat
C:\Windows\System32\DrvConh\mscoree.dll
C:\Windows\System32\DrvConh\ActivationStore.dat
C:\Program Files\Common Files\System\msadc\msadc.inc
C:\ProgramData\desktops.iniThe report includes repeated entries for Dell1.tmp and Dell3.tmp; they are shown once above for readability.
5.3 Domains and command-and-control indicators
| Indicator | Source-reported context |
|---|---|
javadevhouse.co[.]kr | Malicious JavaScript/initial infection infrastructure |
pythonapischool[.]com | Malicious JavaScript/initial infection infrastructure |
hxxps://cubedesign[.]co[.]kr/news | Cubridge C2 |
hxxps://cubedesign[.]co[.]kr/mobile/fashion | Cubridge C2 |
hxxps://fp-news[.]co[.]kr/shop/cart | Cubridge C2 |
hxxps://k-wbc[.]co[.]kr/skin/social/social_view.php | ClientHub C2 |
hxxps://rcsh[.]co[.]kr/modules/bbs/tiles/bbs_div/inputForm.php | ClientHub C2 |
hxxps://www[.]ksog[.]org/insert_file/handle.php | ClientHub C2 |
IOC handling recommendations:
Correlate indicators with endpoint, DNS, proxy, firewall, and network-flow logs.
Review suspicious process relationships involving
fsquirt.exe,INISAFECrossWebEXSvc.exe, and unexpected DLL loading.Investigate unexpected files or services associated with the listed paths.
Check for unusual GitHub API or Dropbox API activity originating from endpoints that do not require those services.
Treat a single IOC match as a lead requiring investigation, not conclusive proof of compromise.
Validate indicator status before deploying blocks, especially where legitimate services or compromised third-party websites may be involved.
6. Potential Impact
6.1 Corporate espionage
The campaign's targeting of software and technology-related organizations creates potential exposure of proprietary information, source code, development environments, and internal systems.
6.2 Credential and browser-data theft
The documented browser-stealing functionality could expose stored credentials, cookies, bookmarks, browsing history, and other sensitive information.
6.3 Long-term access
The multi-stage chain and service-based persistence mechanisms could enable continued access to compromised endpoints and facilitate further payload deployment.
6.4 Security monitoring evasion
In-memory execution, DLL side-loading, timestamp manipulation, and concealed payload delivery can make detection and forensic reconstruction more difficult.
6.5 Broader organizational compromise
If compromised endpoints have privileged access or trusted relationships with other systems, attackers may be able to extend their operations beyond the initial device. The extent of any such lateral movement must be established from incident evidence.
7. Recommendations
7.1 Immediate Actions
Search for published IOCs: Check the listed hashes, paths, domains, and URLs against endpoint, DNS, proxy, and firewall telemetry.
Inspect suspicious Windows services: Review newly created or unusual services, especially those running from unexpected directories or using names that resemble legitimate Windows services.
Examine DLL loading activity: Investigate suspicious DLLs named
mscoree.dlland unusual loading relationships involving legitimate executables such asImeBroker.exe.Review CrossWeb EX environments: Assess installations of INISAFE CrossWeb EX and related financial-security software for suspicious process activity, unexpected child processes, and unapproved modifications.
Investigate memory-resident activity: Use appropriate endpoint detection and response tools to identify suspicious shellcode execution, reflective loading, and unusual process injection.
Audit outbound communications: Look for unexpected GitHub API requests, access to the listed domains, and unusual Dropbox API traffic from endpoints that do not normally use these services.
Protect credentials: If compromise is suspected, revoke or rotate exposed credentials, invalidate sessions, and review account activity for unauthorized access.
Preserve forensic evidence: Collect endpoint logs, process trees, service configuration, memory evidence where available, and relevant network records before rebuilding or reimaging affected systems.
Contain confirmed infections: Isolate affected endpoints and follow the organization's incident response procedures to prevent further access or payload delivery.
7.2 Preventive Actions
Reduce software attack surface: Review mandatory or widely deployed security software for unnecessary privileges, exposed interfaces, and outdated versions.
Apply security updates: Keep Windows, browsers, financial-security applications, and endpoint protection products updated according to vendor and organizational guidance.
Strengthen application control: Use application allowlisting and code-integrity controls where appropriate to limit unauthorized DLL loading and execution from writable directories.
Harden Windows services: Restrict service creation to authorized administrators and monitor new services, unusual executable paths, and unexpected privilege assignments.
Improve memory-threat detection: Configure endpoint monitoring to detect suspicious process injection, shellcode execution, reflective DLL loading, and abnormal process relationships.
Monitor legitimate platforms for abuse: Apply appropriate controls and logging for GitHub and Dropbox access. Use application-aware detection rather than assuming all traffic to these platforms is benign or malicious.
Strengthen phishing defenses: Filter suspicious links, inspect newly registered or unusual domains, and train users to report unexpected résumé, portfolio, and job-related links.
Validate third-party software security: Maintain an inventory of software required to access banking, government, and other sensitive services. Review vendor advisories and remove or disable components that are no longer required.
Protect source code and development systems: Enforce least privilege, multi-factor authentication, repository access controls, and monitoring for unusual source-code access or bulk downloads.
Use layered detection: Correlate identity, endpoint, network, and application events to detect multi-stage activity rather than relying solely on static signatures.
8. MITRE ATT&CK Mapping
The following are suggested ATT&CK technique mappings based on behaviors described in Kaspersky's report. They are editorial mappings for ThreatBeaconX, not a claim that Kaspersky explicitly assigned these technique IDs.
| Technique ID | Technique | Relevance to the reported activity |
|---|---|---|
T1566.002 | Phishing: Spearphishing Link | Targeted messages reportedly directed victims to fabricated portfolio and résumé websites. |
T1189 | Drive-by Compromise | Relevant to the suspected watering-hole approach, although the exact infection trigger was not conclusively determined. |
T1574.002 | Hijack Execution Flow: DLL Side-Loading | TouchHub and TouchShift used DLL side-loading in the documented infection chain. |
T1055 | Process Injection | The infection chain involved shellcode injection and additional in-memory payload-loading techniques. |
T1053 | Scheduled Task/Job | Relevant only where scheduled execution is confirmed; the report describes scheduled execution controls but does not establish that every sample used a Windows scheduled task. |
T1543.003 | Create or Modify System Process: Windows Service | TouchHub registered and started a Windows service to launch the TouchShift payload. |
T1071.001 | Application Layer Protocol: Web Protocols | Malware components used HTTP/HTTPS-style communications and browser-like request characteristics. |
T1102 | Web Service | TechHub used GitHub as a command-and-control communication channel; a ClientHub variant used Dropbox. |
T1027 | Obfuscated Files or Information | Encrypted payloads and concealed in-memory execution were observed in the chain. |
T1005 | Data from Local System | File collection, host profiling, and browser-data collection capabilities were documented. |
T1113 | Screen Capture | Cubridge's function module included screenshot-capture capability. |
These mappings should be validated against the current MITRE ATT&CK knowledge base before publication as formal framework tags.
9. Conclusion
Operation FlashHole demonstrates how the Lazarus Group continues to refine established malware families and combine them into increasingly modular attack chains. By exploiting a trusted software environment, deploying multiple malware stages, and using memory-resident payloads and legitimate online services for communications, the operators sought to establish access while reducing visible traces.
The campaign's use of TechHub, Cubridge, TouchHub, TouchShift, ClientHub, and LPEClient reflects a layered architecture in which individual components handle distinct tasks, from initial payload delivery and persistence to host profiling and command execution.
Kaspersky identified at least three affected South Korean organizations and assessed the operation as part of the broader Cascade Hole campaign. The researchers characterized FlashHole as espionage-focused, while acknowledging that the overall threat landscape includes other Lazarus-linked activity with different objectives.
For defenders, the key lessons are to scrutinize trusted software execution paths, monitor suspicious DLL loading and Windows service creation, investigate unusual traffic to legitimate online platforms, and maintain strong endpoint visibility. Effective detection requires correlating multiple behavioral signals rather than depending on a single file hash or domain.
