Trigram Campaign: Industrialized North Korean-Linked Cyber Operations
Detection stack
- AIDR
- Alert
- ETL
- Query
Summary
The Trigram campaign is an industrialized North Korean-linked operation targeting developers across the AI, blockchain, and logistics sectors. Attackers rely on fake job opportunities and ClickFix social engineering to persuade victims to execute malicious commands that deploy an infostealer. The operation is coordinated through a sophisticated back-office platform that automates targeting, email delivery, and real-time victim monitoring.
Investigation
Stormshield’s CTI team correlated findings from Proofpoint, SafeDep, and Genians to connect the Trigram campaign with the Kimsuky group (APT-43). The investigation uncovered a three-layer infrastructure consisting of decoy recruitment and logistics websites, C2 servers using GitLab as a dead-drop resolver, and a Django-based campaign management platform. Analysts also observed signs of AI-assisted development within the attacker’s operational tooling.
Mitigation
Organizations should enforce strict controls over command execution in terminal environments and train developers to recognize the risks of ClickFix social engineering. Monitoring for unauthorized Node.js installations and suspicious PowerShell or Bash activity is recommended. Security teams should also detect suspicious domain registrations combining brand names with recruitment-related keywords.
Response
When Trigram activity is detected, isolate affected systems to prevent credential theft and cryptocurrency exfiltration. Review terminal history and browser activity to identify the original recruitment lure. Perform a comprehensive audit of developer environments for unauthorized portable Node.js binaries, malicious scripts, and persistent loaders.
Attack Flow
We are still updating this part.
Detections
Nohup Usage (via cmdline)
Possible Execution by Use of Nohup (via cmdline)
Suspicious CURL Usage (via cmdline)
Possible Vscode Automatic Tasks Configuration File Created In Unusual Directory (via file_event)
Suspicious Command and Control by Unusual Top Level Domain (TLD) DNS Request (via dns)
Possible Base64 Encoded Strings Manipulation [MacOS] (via cmdline)
Possible Vscode Automatic Tasks Configuration File Created In Unusual Directory [MACOS] (via file_event)
Possible Base64 Encoded Strings Manipulation (via cmdline)
Remote File Upload / Download via Standard Tools (via cmdline)
Possible Vscode Automatic Tasks Configuration File Created In Unusual Directory [LINUX] (via file_event)
IOCs (HashSha256) to detect: Trigram Campaign
IOCs (HashMd5) to detect: Trigram Campaign
IOCs (SourceIP) to detect: Trigram Campaign
IOCs (DestinationIP) to detect: Trigram Campaign
Trigram Campaign File Download and Command Execution Detection [Windows Sysmon]
Trigram Campaign Command Execution via Clipboard Substitution [Windows Process Creation]
Trigram Campaign Bash Command and Curl File Download [Linux Process Creation]
Detection of Malicious PowerShell and CMD Commands in Trigram Campaign [Windows Powershell]
Simulation Execution
-
Attack Narrative & Commands: An adversary is performing reconnaissance and initial access. They use
curl.exeto download a payload namedverify_wfrom a remote C2 server to bypass basic filename filters. Once the download is complete, they execute a command designed to appear as if a tool has copied a command line (simulating thecmdline_copiedartifact required by the detection logic) to trigger or test the organization’s detection capabilities. -
Regression Test Script:
# Simulation of Trigram Campaign TTPs # 1. Simulate the Proxy Log entry via a curl request # Note: We use -A to force the 'curl' user agent as per detection logic curl.exe -A "curl" "http://localhost/verify_w" # 2. Simulate the Windows Event 4688 with the specific string required # We use cmd.exe to execute a command containing the target string cmd.exe /c "echo cmdline_copied > C:WindowsTempsim_log.txt" -
Cleanup Commands:
# Remove simulated artifacts Remove-Item -Path "C:WindowsTempsim_log.txt" -ErrorAction SilentlyContinue