# Trigram Campaign Impersonates ~50 AI and Blockchain Brands to Deliver Infostealers to Developers (Overlord RAT, install_guard.js; UNK_DeadDrop continuation)

> Trigram is an industrialized campaign targeting AI, blockchain and logistics developers in Europe and the US with ClickFix-style lures (fake recruitment, conference and package-tracking sites impersonating ~50 brands), a forged-commit compromise of the antd-img-crop GitHub repository, and trojanized repositories, all delivering the install_guard.js infostealer and the Overlord RAT. Stormshield assesses with high confidence that it continues Proofpoint's UNK_DeadDrop cluster; North Korea-aligned similarities (Contagious Interview, Kimsuky) are noted but attribution is not proven.

- **Published:** 2026-11-20T06:12:29Z
- **Last reviewed:** 2026-11-20T06:12:29Z
- **Canonical:** https://intel.threadlinqs.com/threat/TL-2026-6548
- **Card image:** https://intel.threadlinqs.com/og/threat/TL-2026-6548.png
- **ID:** TL-2026-6548
- **Severity:** HIGH
- **Category:** MALWARE
- **Status:** ACTIVE
- **Actor:** UNK_DeadDrop (North Korea)
- **Detections:** 9 (1 shown free below) · **IOCs:** 29 (3 shown free below)

## Description

Stormshield's CTI team (published 2026-11-17) describes Trigram as a three-layer operation: (1) decoy sites impersonating about 50 Web3/AI brands (recruitment lures on domains such as join*, buildwith*, inv2*, *hires, *careers, *talents, *council), a fake ICAGLLMBT conference site, and DHL/DPD/Aramex package-tracking pages; (2) C2/relay infrastructure that serves install_guard.js and receives installation telemetry; and (3) an operator back office called rms-platform (Django, exposed in debug mode on 170.205.30.27:8000) that scrapes GitHub profiles to build target lists, sends phishing mail through Resend, Brevo, Mailgun and Smartlead, issues referral codes and waybill codes bound to the victim IP, records visitor sessions with FullStory, scores targets as important/moderate/useless, exports data from an external MongoDB, and includes a wallet monitoring module. Stormshield rates the three lure scenarios as one operator/one platform with high confidence.

Infection chain: the victim passes a referral-code gate, is told to paste a 'human verification' command into a terminal that merely POSTs a nonce to /api/verify, sees a deliberate 'Operation timeout... Nonce Refreshed' error, and from then on the Copy button and Ctrl+C silently place a different command (with trailing newline) on the clipboard. That command curls (curl -ks) a stage-2 loader: v.cmd (Windows, from /verify_w), u or verify_m (Linux/macOS, from /verify_m). The loader decodes an embedded Base64 script, takes a lock, checks for Node.js >= 18 and otherwise downloads portable Node.js 20.18.1, then fetches install_guard.js (must be >= 50 KB) and runs it. The kit presents itself as a 'deepfake_guard' tool and reports each step (copiedCmdline through passwordSaved) to guard-launcher-result, guard-install-log and guard-install-progress endpoints. On Windows the loader tries to remove Mark-of-the-Web (Zone.Identifier).

SafeDep (2026-11-03) showed the same payload chain delivered through the nanxiaobei/antd-img-crop GitHub repository (about 98,000 npm downloads/week; the npm package itself was not affected): on 2026-10-31 at 15:12 UTC an unsigned forged '4.30.0' commit (2aa850469fa1ff2e2efe8eebdf1e085e4d11212a, reusing the real March 12 author/date) was force-pushed, adding a prebuild hook (scripts/prebuild.js) that fetches verify_m/verify_w from icagllmbt.space when a developer runs npm run build. Stage 3, install_guard.js (AES-256-GCM encrypted, loaded in memory), resolves its C2 through a GitLab dead-drop (gitlab.com/drop-indexings/tasks, later up-indexings/tasks, config.json listing the current API domain), steals browser credentials/cookies, macOS Keychain and Linux keyring contents, browser-extension and desktop crypto wallets, .env/.npmrc/.gitconfig/SSH/KeePass files, and installs the OverLord agent (Go module overlord-client; Node variant on Windows) disguised as Google Update Supporter with persistence via LaunchDaemon com.google.UpdateSupporter, systemd google-agent.service, XDG autostart and the scheduled task GoogleUpdateSupporterNode. On macOS and Linux a fake system dialog harvests the user's password; on Windows a short-lived SYSTEM scheduled task is used to decrypt Chromium keys. A sideloaded VS Code/Cursor extension named google-dev-tools.google-update-support is another entry vector (SafeDep's assessment, inferred rather than confirmed).

Attribution: Stormshield assesses with high confidence a continuation of UNK_DeadDrop (Proofpoint, 2026-06-08: trojanized repos with .vscode/tasks.json folderOpen tasks, Overlord-derived google-update-support agents, asteara.org and 170.205.29.83 infrastructure), noting an evolution to multiple initial-access vectors and remote payload fetching. Operator identities in the back-office code and repos (eros1030109@gmail.com as hard-coded super-admin, 0xapp123, silviattech1129, alexlopez-g, harper.danie95@gmail.com) link to a Genians Kimsuky report (2026-08-10) and OctoProbe's GitPower research. Proofpoint rates UNK_DeadDrop 'very likely North Korea-aligned' but unconfirmed, SafeDep explicitly says technique overlap with Contagious Interview does not prove the operators, and Stormshield does not state a firm Kimsuky link, so attribution confidence is LOW. Stormshield promised a follow-up article on Overlord and the wallet-theft details. No CVE is involved and no CVSS applies. Severity HIGH is an analyst judgment.

## MITRE ATT&CK

- T1593.003 Code Repositories
- T1583.001 Domains
- T1583.006 Web Services
- T1566.002 Spearphishing Link
- T1195.002 Compromise Software Supply Chain
- T1204.004 Malicious Copy and Paste
- T1059.007 JavaScript
- T1059.004 Unix Shell
- T1059.001 PowerShell
- T1176.002 IDE Extensions
- T1543.004 Launch Daemon
- T1053.005 Scheduled Task
- T1553.005 Mark-of-the-Web Bypass
- T1036.005 Match Legitimate Resource Name or Location
- T1070.004 File Deletion
- T1555.003 Credentials from Web Browsers
- T1555.001 Keychain
- T1056.002 GUI Input Capture
- T1552.001 Credentials In Files
- T1102.001 Dead Drop Resolver
- T1219 Remote Access Tools

## Free detection (Splunk SPL, Microsoft KQL, Sigma)

**Trigram ClickFix Stage-2 Loader Fetch (verify_m / verify_w / icagllmbt.space) — Exploitation** (`TL-2026-6548-SPL-001`)

Detects the pasted ClickFix terminal command used by the Trigram (UNK_DeadDrop continuation) campaign: curl -ks retrieval of the stage-2 loader (verify_m on Linux/macOS, verify_w/v.cmd on Windows) from icagllmbt.space, plus follow-on references to install_guard.js, the deepfake_guard work directory, or the api.gaganata.ink / api.fallgganata.ink stage-3 stores. Catches the user-execution moment before the Node.js infostealer and OverLord agent are installed.

- **Languages:** Splunk SPL, Microsoft KQL, Sigma
- **ATT&CK:** T1204.004, T1059.004, T1059.001, T1105
- **Data sources:** sysmon, linux_auditd, defender_for_endpoint
- **Severity:** high
- **Confidence:** high

### Splunk SPL

````spl
(index=sysmon sourcetype="XmlWinEventLog:Microsoft-Windows-Sysmon/Operational" EventCode=1 (CommandLine="*icagllmbt.space*" OR CommandLine="*/verify_w*" OR CommandLine="*/verify_m*" OR CommandLine="*install_guard.js*" OR CommandLine="*deepfake_guard*" OR CommandLine="*api.gaganata.ink*" OR CommandLine="*api.fallgganata.ink*")) OR (index=linux sourcetype="linux:audit" type=EXECVE ("icagllmbt.space" OR "verify_m" OR "install_guard.js" OR "deepfake_guard")) | eval cmd=coalesce(CommandLine,_raw), img=coalesce(Image,a0), actor=coalesce(User,auid) ```Trigram ClickFix: curl -ks to verify_m/verify_w on icagllmbt.space then install_guard.js``` | eval stage=case(like(cmd,"%icagllmbt.space%"),"stage2_host_icagllmbt", like(cmd,"%install_guard.js%"),"stage3_install_guard_js", like(cmd,"%deepfake_guard%"),"deepfake_guard_workdir", true(),"stage2_loader_verify_path") | stats count min(_time) as first_seen max(_time) as last_seen values(stage) as stage values(ParentImage) as parent_image values(cmd) as command_line by host actor img | convert ctime(first_seen) ctime(last_seen) | table first_seen last_seen host actor img parent_image stage command_line count
````

### Microsoft KQL

```kql
// Trigram ClickFix stage-2 loader fetch (curl -ks / PowerShell) - TL-2026-6548
let c2_hosts = dynamic(["icagllmbt.space","api.gaganata.ink","api.fallgganata.ink","api.chestsoi.ink","api.cerysoi.ink","api.gamsterb.ink","asteara.org"]);
DeviceProcessEvents
| where Timestamp > ago(24h)
| where ProcessCommandLine has_any (c2_hosts)
    or (ProcessCommandLine has_any ("/verify_w","/verify_m") and FileName in~ ("curl","curl.exe","wget","powershell.exe","pwsh.exe","cmd.exe","bash","sh","zsh"))
    or ProcessCommandLine has_any ("install_guard.js","deepfake_guard","nvdg_worker.vbs")
| extend CurlInsecureSilent = ProcessCommandLine has_any ("curl -ks","curl -sk")
| project Timestamp, DeviceName, AccountName, FileName, ProcessCommandLine, CurlInsecureSilent, InitiatingProcessFileName, InitiatingProcessCommandLine, SHA256
| order by Timestamp desc
```

### Sigma

```yaml
title: Trigram ClickFix Stage-2 Loader Fetch via curl or wget
id: 3a7d9c10-5e42-4f8b-b1a6-2d94c7e08f11
status: experimental
description: Detects curl/wget retrieving the Trigram verify_m/verify_w stage-2 loader, the icagllmbt.space host or install_guard.js, as pasted by victims during the ClickFix referral-code lure.
references:
    - https://www.stormshield.com/news/cti-trigram-campaign/
    - https://safedep.io/blog/antd-img-crop-forged-commit-overlord-stealer
author: Threadlinqs Team
date: 2026-11-20
tags:
    - attack.execution
    - attack.t1204.004
    - attack.t1059.004
    - attack.command-and-control
    - attack.t1105
    - attack.tool.overlord
logsource:
    category: process_creation
    product: linux
detection:
    selection_loader:
        CommandLine|contains:
            - 'icagllmbt.space'
            - 'install_guard.js'
            - 'deepfake_guard_run.sh'
            - 'deepfake_guard_work'
    selection_curl:
        Image|endswith:
            - '/curl'
            - '/wget'
        CommandLine|contains:
            - '/verify_m'
            - '/verify_w'
    condition: 1 of selection_*
falsepositives:
    - Security researchers or sandboxes deliberately fetching the Trigram loader
    - Internal tooling that happens to use a /verify_m or /verify_w path
level: high

```

**Known false positives:**

- Security researchers or sandboxes deliberately detonating the Trigram loader
- Internal tooling that uses a /verify_m or /verify_w URL path on a different host

**Tuning notes:** Highest fidelity is the icagllmbt.space and install_guard.js strings; the bare /verify_m and /verify_w paths are generic, so require them alongside curl/wget or add the C2 hosts when reducing noise. Refresh the host list from the threat IOC set as the operators rotate domains. Ensure Linux auditd EXECVE logging is enabled for the linux branch.

### 8 more detection rules (Splunk SPL, Microsoft KQL, Sigma), unlock with TL Creds

- Trigram [indicator] Prebuild Hook Dropper Fetching Stage-2 Loader — Exploitation (locked: unlock with TL Creds)
- Trigram Fake Password Dialog and Keychain / Secret File Harvesting on macOS and Linux — Post-Exploitation (locked: unlock with TL Creds)
- Trigram Node.js GitLab Dead-Drop Resolver Followed by .ink C2 Connection — C2 (locked: unlock with TL Creds)
- Trigram Sideloaded IDE Extension google-dev-tools.google-update-support and Portable Node — Exploitation (locked: unlock with TL Creds)
- Trigram OverLord LaunchDaemon and systemd Persistence Disguised as Google Update Supporter — Post-Exploitation (locked: unlock with TL Creds)
- Trigram and UNK_DeadDrop Infrastructure IP Connections (rms-platform, lure hosting, C2 5173) — C2 (locked: unlock with TL Creds)
- Trigram OverLord Windows Persistence and Staging Artifacts (GoogleUpdateSupporterNode) — Post-Exploitation (locked: unlock with TL Creds)
- Trigram C2 and Relay Domain Resolution (gaganata.ink, chestsoi.ink, cerysoi.ink, [indicator]) — C2 (locked: unlock with TL Creds)

Sign up free to unlock the detection rules: https://threadlinqs.com/?modal=signup

## Sample indicators (3 of 29)

Defanged, TLP:CLEAR.

- IP address: `170.205.30[.]27`
- Domain: `asteara[.]org`
- URL: `hxxps://gitlab[.]com/drop-indexings/tasks/-/raw/main/config.json`

Full set of 29 indicators unlocks with TL Creds: https://threadlinqs.com/tl-creds

## Sources

- [Stormshield CTI: Trigram campaign](https://www.stormshield.com/news/cti-trigram-campaign/)
- [The Hacker News: Trigram Campaign Impersonates About 50 AI and Blockchain Brands](https://thehackernews.com/2026/11/trigram-campaign-impersonates-about-50.html)
- [SafeDep: antd-img-crop GitHub Repository Hijacked to Deliver OverLord Stealer](https://safedep.io/blog/antd-img-crop-forged-commit-overlord-stealer)
- [SafeDep IOC list (CSV): antd-img-crop / OverLord](https://safedep.io/data/blogs/antd-img-crop-overlord-iocs.csv)
- [Proofpoint: Don't Fear the Repo - UNK_DeadDrop phishing campaign targets developers](https://www.proofpoint.com/us/blog/threat-insight/dont-fear-repo-unkdeaddrop-phishing-campaign-targets-developers-steal)
- [Genians: Kimsuky integrates AI into attack operations (Operation GitPower)](https://www.genians.co.kr/en/blog/threat_intelligence/kimsuky_ai_llm)
- [Genians: Operation GitPower (Ukraine targeting)](https://www.genians.co.kr/en/blog/threat_intelligence/operation-gitpower)
- [OctoProbe: Kimsuky Operation GitPower GitHub C2](https://www.octoprobe.ai/blog/kimsuky-operation-gitpower-github-c2/)
- [Contra community: alert on fake designer job offers](https://contra.com/community/weG4gy2L-alert-avoid-fake-designer-job-offers)
- [FBI IC3 alert on Contagious Interview (cited by Stormshield)](https://www.ic3.gov/CSA/2026/260918.pdf)
- [DailySecurityReview: North Korea-Linked Trigram Targets Web3 Developers With ClickFix](https://dailysecurityreview.com/?p=45809)

## Full data

The other 8 detection rules (Splunk SPL / Microsoft KQL / Sigma) unlock with TL Creds; a free account includes daily creds: https://threadlinqs.com/?modal=signup
Indicator values unlock with TL Creds, at a lower price on Red and Purple; AI agents can query rules and IOCs through the Threadlinqs MCP server (Purple and Gold in full, 2 calls a day on a free account): https://intel.threadlinqs.com/mcp

Canonical: https://intel.threadlinqs.com/threat/TL-2026-6548
