Skip to navigation

Threat reportSupply ChainTL-2026-0559

DPRK npm Supply Chain Attack — terminal-logger-utils Abuses Hugging Face (Lordplay/system-releases) to Deliver Node.js SEA Keylogger/Infostealer/RAT

highACTIVE

North Korea-linked npm uploader jpeek895 (and three colluding accounts) published terminal-logger-utils plus three dependents (pretty-logger-utils, ts-logger-pack, pinno-loggers) whose postinstall hook executes utils.cjs, an obfuscated multi-stage dropper. utils.cjs fingerprints the host OS, pulls a Node.js Single Executable Application (SEA) from the attacker-controlled Hugging Face repository Lordplay/system-releases, and establishes triple persistence on Windows (%LOCALAPPDATA%\MicrosoftSystem64 + hidden VBS launcher + scheduled task + HKCU\...\Run key).

Key facts

IdentifiersTL-2026-0559
Also known asterminal-logger-utils campaign, Lordplay/system-releases campaign, jpeek895 npm cluster
SeverityHIGH
Affected productsnpm registry (OpenJS / GitHub) terminal-logger-utils; npm registry (OpenJS / GitHub) pretty-logger-utils; npm registry (OpenJS / GitHub) ts-logger-pack; npm registry (OpenJS / GitHub) pinno-loggers; Any Node.js developer workstation Windows, macOS, Linux (OS-dispatched payloads)
Weaknesses (CWE)CWE-506, CWE-829, CWE-494, CWE-1357
AttributionContagious Interview (high confidence)
Nation-state nexusNorth Korea
Malwareterminal-logger-utils SEA implant
ToolsCustom Node.js SEA implant (keylogger+infostealer+RAT) over WebSocket, Node.js Single Executable Application (SEA)
MITRE ATT&CK techniquesT1005 Data from Local System, T1027 Obfuscated Files or Information, T1036 Masquerading, T1041 Exfiltration Over C2 Channel, T1053 Scheduled Task/Job, T1056 Input Capture, T1059 Command and Scripting Interpreter, T1071 Application Layer Protocol, T1082 System Information Discovery, T1083 File and Directory Discovery, T1095 Non-Application Layer Protocol, T1102 Web Service and 14 more in the attack flow
MotivationFINANCIAL
Target sectorstechnology, software-development, cryptocurrency, financial, fintech, web3, ai-ml, devops, cloud, open-source
Target regionsGlobal, North America, Europe, Asia-Pacific
Detection rules9 (1 shown free)
Sources7 cited sources

Detection coverage

As of 2026-05-22, Threadlinqs Intelligence publishes 9 detection rule(s) for TL-2026-0559 across Splunk SPL, Microsoft KQL and Sigma, covering 26 indicator(s) of compromise. The whole corpus is readable without an account. Every threat comes with one complete detection in Splunk SPL, Microsoft KQL and Sigma, free for everyone; a free account adds daily TL Creds to unlock the rest. Paid plans add a larger daily allowance, correlation and the MCP server. Threadlinqs MCP server · View plans.

Free detection: DPRK terminal-logger-utils — WebSocket C2 195.201.194.107 + Hugging Face Lordplay/system-releases + Keystroke Exfil (C2)

Detects the C2/exfiltration phase: outbound connections to the Hetzner WebSocket C2 IP 195.201.194.107 (T1071.001/T1102.002), HTTPS requests to huggingface.co for the attacker repo path /Lordplay/system-releases (T1102/T1105), or HTTP POSTs whose URI path contains /api/validate/keyboard-events — the bespoke keystroke exfil endpoint (T1041/T1056.001).

  • Languages: Splunk SPL, Microsoft KQL, Sigma
  • ATT&CK: T1071.001, T1102, T1102.002, T1105, T1041, T1056.001
  • Data sources: proxy_logs, firewall_logs, zeek_conn, zeek_http, edr_network_telemetry
  • Severity: critical
  • Confidence: high

Splunk SPL

(index=proxy (dest_ip="195.201.194.107" OR uri_path="*api/validate/keyboard-events*" OR (host="huggingface.co" uri_path="*Lordplay/system-releases*"))) OR (index=firewall dest_ip="195.201.194.107") OR (index=zeek sourcetype="zeek:conn" id.resp_h="195.201.194.107") | eval c2_vector=case(match(uri_path,"api/validate/keyboard-events"),"keystroke_exfil_uri", (host="huggingface.co" AND match(uri_path,"Lordplay/system-releases")),"huggingface_payload_pull", (dest_ip="195.201.194.107" OR id.resp_h="195.201.194.107"),"hetzner_ws_c2", 1=1,"unknown") | eval threat_id="TL-2026-0559", technique="T1071.001,T1102.002,T1102,T1105,T1041,T1056.001" | stats count min(_time) as first_seen max(_time) as last_seen values(src_ip) as sources values(host) as hosts values(uri_path) as paths values(c2_vector) as vectors by dest_ip | sort - count

Microsoft KQL

DeviceNetworkEvents | where Timestamp > ago(24h) | where RemoteIP == "195.201.194.107" or (RemoteUrl has "huggingface.co" and RemoteUrl has "Lordplay/system-releases") or RemoteUrl has "/api/validate/keyboard-events" | extend C2Vector = case(RemoteUrl has "/api/validate/keyboard-events", "keystroke_exfil_uri", RemoteUrl has "Lordplay/system-releases", "huggingface_payload_pull", RemoteIP == "195.201.194.107", "hetzner_ws_c2", "unknown") | project Timestamp, DeviceName, RemoteIP, RemotePort, RemoteUrl, InitiatingProcessFileName, InitiatingProcessCommandLine, C2Vector | extend ThreatID="TL-2026-0559"

Sigma

title: DPRK terminal-logger-utils C2 Network Indicators (TL-2026-0559)
id: 7f1e0a5a-2026-0559-1003
status: experimental
description: Detects C2 / exfiltration network indicators — 195.201.194.107, Hugging Face Lordplay/system-releases, /api/validate/keyboard-events.
references:
  - https://cybersecuritynews.com/hackers-use-hugging-face/
  - https://attack.mitre.org/techniques/T1102/
author: Threadlinqs Team
date: 2026-05-22
tags:
  - attack.command_and_control
  - attack.t1071.001
  - attack.t1102
  - attack.t1102.002
  - attack.t1105
  - attack.exfiltration
  - attack.t1041
  - attack.collection
  - attack.t1056.001
logsource:
  category: proxy
detection:
  c2_ip:
    dst_ip: '195.201.194.107'
  hf_payload:
    cs-host: 'huggingface.co'
    cs-uri-stem|contains: '/Lordplay/system-releases'
  keylog_uri:
    cs-method: POST
    cs-uri-stem|contains: '/api/validate/keyboard-events'
  condition: c2_ip or hf_payload or keylog_uri
falsepositives:
  - Legitimate Hugging Face traffic excluded by requiring the attacker-controlled Lordplay/system-releases path
  - Other Hetzner customers excluded by exact IP match
level: critical

Known false positives:

  • Legitimate huggingface.co traffic — narrowed out by attacker-controlled Lordplay/system-releases path
  • Other Hetzner-hosted services — narrowed out by exact IP 195.201.194.107

Tuning notes: Promote any 195.201.194.107 hit to incident automatically — no benign traffic expected. The /api/validate/keyboard-events URI is bespoke and high-fidelity regardless of destination host — never narrow it by host. Retain Lordplay path rule for retroactive hunting after takedown.

8 more detection rules (Splunk SPL, Microsoft KQL, Sigma) · unlock with TL Creds. Sign up free to unlock the detection rules or query them with the Threadlinqs MCP server (Purple)
  • DPRK [indicator] — node.exe Postinstall Spawning Shell/WScript on Defender Endpoint (Exploitation) · unlock with TL Creds
  • DPRK [indicator] — MicrosoftSystem64 File Drops + Run Key + Scheduled Task (Post-Exploitation) · unlock with TL Creds
  • DPRK [indicator] — Defender Network: [indicator] + Hugging Face Lordplay/system-releases + Keystroke URI (C2) · unlock with TL Creds
  • DPRK [indicator] — Hidden VBS Launcher in MicrosoftSystem64 Masquerade Path (Exploitation->Execution) · unlock with TL Creds
  • DPRK [indicator] — Linux User Crontab @reboot + systemd --user in microsoft-system64 Masquerade (Post-Exploitation) · unlock with TL Creds
  • DPRK [indicator] — WebSocket Upgrade w/ Host [indicator] + Hugging Face Lordplay/system-releases Pull (C2) · unlock with TL Creds
  • DPRK [indicator] — npm postinstall (node.exe) Spawning Shell or WScript Child (Exploitation) · unlock with TL Creds
  • DPRK [indicator] — Windows Triple Persistence in MicrosoftSystem64 Masquerade (Post-Exploitation) · unlock with TL Creds

Indicators of compromise

As of 2026-05-22, Threadlinqs Intelligence tracks 26 indicators of compromise for TL-2026-0559.

Sample indicators (2 of 26)

  • IP address: 195.201.194[.]107
  • Domain: huggingface[.]co

Defanged, TLP:CLEAR. Full set of 26 indicators: unlock with TL Creds.

26 indicators of compromise · unlock with TL Creds. Compare plans

Technical analysis

The implant fuses keylogger, infostealer, and RAT functionality — exfiltrating Telegram session data, SSH keys, cryptocurrency wallets, browser credentials, cloud configs, and environment variables — with HTTP keystroke exfil to /api/validate/keyboard-events and a WebSocket C2 at 195.201.194.107 for full interactive machine control. The campaign is a novel TTP evolution: abusing Hugging Face — a trusted AI/ML platform — as both payload-hosting CDN and self-update channel to evade allowlist-based egress controls.

On 2026-05-22, OX Security disclosed an active North Korea-aligned npm supply chain attack centered on the package terminal-logger-utils, uploaded by npm account jpeek895 — an account previously flagged by independent researcher kmsec.uk as part of the DPRK npm-package cluster. Three dependent packages — pretty-logger-utils, ts-logger-pack, pinno-loggers — were published by collaborating accounts jpeek886, pvnd3540749, and yggedd817513 to inflate apparent legitimacy and broaden the install footprint.

Infection Chain. When a developer runs `npm install terminal-logger-utils` (directly or transitively via one of the dependents), npm's `postinstall` lifecycle hook executes the bundled dropper `utils.cjs`. utils.cjs is a heavily obfuscated CommonJS module that performs OS dispatch (Windows / macOS / Linux), then issues an HTTPS request to the Hugging Face Hub repository `Lordplay/system-releases` (a legitimate-looking AI model hosting endpoint) to download an OS-specific Node.js Single Executable Application (SEA) binary. The SEA payload bundles a Node runtime with the malicious JavaScript, eliminating any requirement for a developer-side Node installation and reducing AV/EDR detection of the script body.

Windows Persistence. On Windows hosts the second-stage installs itself to `%LOCALAPPDATA%\MicrosoftSystem64\` — a path chosen to masquerade as a legitimate Microsoft system component. Three persistence mechanisms are established in parallel: (1) a hidden VBScript launcher dropped into the install directory that re-launches the SEA binary on user logon, (2) a Windows Scheduled Task registered under Task Scheduler that re-executes the binary at logon and on a recurring interval, and (3) an `HKCU\Software\Microsoft\Windows\CurrentVersion\Run` registry key as a fallback autostart vector. The triple-redundant persistence is characteristic of DPRK developer-targeting clusters tracked under names such as Contagious Interview / DEV#POPPER / Famous Chollima.

Implant Capabilities. The SEA second-stage is a fused keylogger + infostealer + RAT. Keystroke capture occurs continuously and is exfiltrated over plain HTTP POST to the path `/api/validate/keyboard-events` on attacker infrastructure. Stealer modules target high-value developer artifacts: Telegram `tdata` session folders (allowing full account hijack without 2FA prompt), `~/.ssh/` keys, cryptocurrency wallet files for major hot-wallet clients (MetaMask, Phantom, Exodus, Electrum, etc.), browser-saved passwords / cookies / autofill (Chrome, Edge, Brave, Firefox profile directories), cloud configuration files (`.aws/credentials`, `gcloud` config, `kube/config`), and shell environment variables that frequently contain API tokens and database URIs. The RAT module opens a WebSocket back to `195.201.194.107` (a Hetzner-range IP) supporting interactive commands: arbitrary shell command execution, file read/write, screenshot capture, simulated input injection, and self-update via fetching new SEA blobs from the Hugging Face repository.

Hugging Face Abuse. The novel element is the abuse of Hugging Face as a malware delivery and update CDN. Hugging Face is allowlisted in many enterprise egress policies because of legitimate ML/AI workloads, and its content-addressable storage backed by Git-LFS provides versioned, high-throughput, TLS-protected delivery. The attackers benefit from: (a) trust inheritance from the platform brand, (b) bypassing URL-reputation-only blocking, (c) free hosting and bandwidth, (d) the ability to push updates by simply git-pushing new SEA blobs to `Lordplay/system-releases`, and (e) plausible deniability — the repo can be styled to look like a legitimate model release. The same pattern was previously observed with the abuse of GitHub Releases, Cloudflare Pages, and Bitbucket; Hugging Face is the next logical evolution.

Attribution. Direct attribution rests on continuity between jpeek895 and prior DPRK npm clusters catalogued by kmsec.uk. The package-naming convention (developer-tool typosquats targeting npm/TypeScript ecosystem users), the multi-account publish-and-prop pattern, the developer-focused exfiltration targets (Telegram for social-engineering pivots, crypto wallets for revenue, cloud credentials for downstream intrusion), and the triple Windows persistence stack all align with the Contagious Interview / Famous Chollima TTP cluster operated by units under the DPRK Reconnaissance General Bureau (RGB) — overlapping with Lazarus subgroup tracking. Attribution confidence: HIGH.

MITRE ATT&CK techniques by tactic:

Collection

T1005 Data from Local System; T1113 Screen Capture

Defense Evasion

T1027 Obfuscated Files or Information; T1036 Masquerading; T1564 Hide Artifacts

Exfiltration

T1041 Exfiltration Over C2 Channel; T1567 Exfiltration Over Web Service

Persistence

T1053 Scheduled Task/Job; T1547 Boot or Logon Autostart Execution

Credential Access

T1056 Input Capture; T1552 Unsecured Credentials; T1555 Credentials from Password Stores

Execution

T1059 Command and Scripting Interpreter; T1204 User Execution

Command and Control

T1071 Application Layer Protocol; T1095 Non-Application Layer Protocol; T1102 Web Service; T1105 Ingress Tool Transfer

Discovery

T1082 System Information Discovery; T1083 File and Directory Discovery

Initial Access

T1195 Supply Chain Compromise

Impact

T1531 Account Access Removal

Resource Development

T1583 Acquire Infrastructure; T1585 Establish Accounts; T1587 Develop Capabilities; T1588 Obtain Capabilities

Remediation

Patches

  • Report and remove the malicious packages from the npm registry (npm-security); ensure transitively pinned versions are purged from organization lockfiles.

Immediate actions

  • Remove terminal-logger-utils, pretty-logger-utils, ts-logger-pack, and pinno-loggers from all package.json and lockfiles; rebuild node_modules from a clean cache.
  • Block egress to 195.201.194.107 at perimeter firewalls and EDR network policy.
  • Block or alert on outbound HTTPS to huggingface.co paths matching Lordplay/system-releases and any unknown user/org repo containing executable binary blobs (.exe, .bin, SEA artifacts).
  • Hunt for %LOCALAPPDATA%\MicrosoftSystem64\ directory on all Windows developer endpoints; quarantine and forensicate any host where it exists.
  • Audit HKCU\Software\Microsoft\Windows\CurrentVersion\Run for entries pointing into MicrosoftSystem64 or referencing .vbs launchers in LocalAppData.
  • Audit Windows Scheduled Tasks for unsigned tasks launching binaries from %LOCALAPPDATA%.
  • Rotate all developer credentials potentially exposed: SSH keys, AWS/GCP/Azure tokens, Telegram sessions, browser-saved passwords, cryptocurrency wallet seeds, npm tokens, GitHub PATs.

Workarounds

  • Pin all npm dependencies to known-good versions in package-lock.json / pnpm-lock.yaml and refuse to upgrade until provenance is verifiable.
  • Run `npm install` only inside ephemeral sandboxed containers; never on host developer machines that hold long-lived credentials.

Longer-term hardening

  • Enforce npm postinstall script disablement in CI and on developer workstations (`npm config set ignore-scripts true`) and re-enable only for vetted dependencies.
  • Adopt an internal npm proxy (Verdaccio, JFrog Artifactory, Sonatype Nexus, GitHub Packages) with allowlist-based publisher verification.
  • Deploy EDR with behavioral detection for: postinstall spawning powershell/cmd, postinstall writing to LocalAppData, scheduled-task creation by node.exe, Run-key writes by node.exe.
  • Egress allowlist Hugging Face only for ML/AI workloads; segregate dev workstations from the ML egress profile or require proxy inspection of large LFS pulls.
  • Adopt Sigstore-style cryptographic signing or npm package provenance attestations as a gating policy for production dependencies.

Affected and fixed versions

  • npm registry (OpenJS / GitHub) — terminal-logger-utils
    Vulnerable versions: all published versions
  • npm registry (OpenJS / GitHub) — pretty-logger-utils
    Vulnerable versions: all published versions
  • npm registry (OpenJS / GitHub) — ts-logger-pack
    Vulnerable versions: all published versions
  • npm registry (OpenJS / GitHub) — pinno-loggers
    Vulnerable versions: all published versions
  • Any Node.js developer workstation — Windows, macOS, Linux (OS-dispatched payloads)
    Vulnerable versions: all

Timeline

  • Hugging Face Hub identified as a new trusted-platform abuse vector for DPRK developer-targeting malware delivery — an evolution beyond prior abuse of GitHub Releases, Bitbucket, and Cloudflare Pages.
  • Campaign confirmed active: malicious packages observed live on the npm registry under the listed accounts; Hugging Face repository Lordplay/system-releases serving SEA binaries; WebSocket C2 locked-ioc-5 reachable.
  • Threadlinqs Intelligence publishes TL-2026-0559 with full MITRE ATT&CK mapping, IOC catalogue, and detection coverage for SOC consumption.
  • Attribution to DPRK established via continuity with prior jpeek895 activity previously documented by independent researcher kmsec.uk.
  • Package list (terminal-logger-utils + 3 dependents), npm uploader accounts (jpeek895, jpeek886, pvnd3540749, yggedd817513), C2 IP locked-ioc-5, and Hugging Face repository Lordplay/system-releases publicly catalogued.
  • OX Security publishes research on the terminal-logger-utils campaign; Cyber Security News covers the disclosure, exposing Hugging Face as the second-stage delivery platform.
  • As of 2026-05-29, this DPRK (Famous Chollima/Contagious Interview) npm supply-chain attack remains active: disclosed only days earlier (2026-05-22 OX Security; JFrog 04-23 on the same Lordplay/system-releases repo), with no confirmed takedown of the packages, Hugging Face CDN, or C2 locked-ioc-5. The actor is escalating, with 204 malicious npm releases Apr 25-May 25, 2026.

Sources cited

More from Contagious Interview

Previous and next in supply chain

Further reading

Drafted by the Threadlinqs research pipeline from 7 sources; published ; last updated —.

Threadlinqs Intelligence — Real-Time Threat Detection Platform

// threat_feed
$ sort --newest
Showing all threats

Live intelligence console

Threat level

Latest Threats