Researchers Identify Three New PhantomRaven npm Attack Waves

· Original article ↗

Summary

Endor Labs identified three PhantomRaven waves from November 2025 to February 2026, distributing 88 npm packages. Researchers say the packages fetch malware remotely to steal developer configuration, CI/CD tokens and system data; 81 reportedly remain available.

Key points

  • Koi first identified PhantomRaven in October 2025 and said it had published 126 malicious packages since August; Endor Labs found three later waves with 88 packages distributed through 50 disposable accounts.
  • Endor Labs reported that 81 malicious packages remained available in the npm registry.
  • The attackers used slopsquatting and LLM-like package names to impersonate projects including Babel and GraphQL Codegen.
  • Remote Dynamic Dependencies in package.json point to external dependencies; running npm install downloads and executes the malware, helping it evade automated package inspection.
  • The malware collects data from .gitconfig, .npmrc and environment variables, targets CI/CD tokens for GitHub, GitLab, Jenkins and CircleCI, and gathers system details.
  • Researchers say the malware exfiltrates data to attacker servers, usually with HTTP GET requests, with HTTP POST and WebSocket also used.
  • Endor Labs observed consistent infrastructure across four waves and nearly unchanged payload code, alongside attackers rotating accounts, package metadata and PHP endpoints.

Article Details

Attack Vectors
  • PhantomRaven used slopsquatting and LLM-like package names to publish malicious packages impersonating established projects on npm.
  • The packages used Remote Dynamic Dependencies in package.json to fetch malware from external servers when a developer ran npm install.
  • The malware collected configuration data, environment variables, and CI/CD tokens, then sent the stolen data to attacker C2 servers using HTTP GET, HTTP POST, or WebSocket.
Defensive Notes
  • Verify the legitimacy of software dependencies and use packages from reputable publishers.
  • Avoid copying package suggestions from AI chatbots or other unvetted sources.

MITRE ATT&CK

Threat Actors

Products

Related Articles