GlassWorm Compromises OVH CLI as GhostAction Adopts Fake-Font Malware

· Original article ↗

Summary

Researchers discuss GlassWorm's compromise of four OVH CLI versions, malware split across packages, and GhostAction's use of fake-font credential stealers that persist beyond removal of malicious GitHub workflows.

Key points

  • Researchers attributed four compromised OVH CLI versions to GlassWorm, linking the contributor account to malicious repository pushes on March 5, seven months earlier. The credential-theft date remains unknown.
  • The malicious CLI builds reportedly came from a contributor's laptop rather than the normal CI pipeline. A user noticed suspicious behavior; researchers later reported lingering malicious pre-install scripts despite announced incident response.
  • Compound payloads distribute functionality across packages to evade per-package detection. Researchers identified webrunc as a possible component, not a package containing a complete malicious payload.
  • Researchers described Contagious Interview delivery through private repositories whose .npmrc configuration selects an alternative registry, allowing malicious dependencies to resolve within the target's project environment.
  • Git forensics linked GhostAction's malicious workflows to subsequent fake-font payload additions. Shared attacker infrastructure and the repository timeline supported attribution despite similarities to PolinRider.
  • The fake-font JavaScript targets developer credentials and SSH private keys, then installs a cron job that checks in every five minutes. Researchers said the stealer only partly works and found one maintainer had removed malicious workflows but missed this payload.
  • The hosts recommended auditing CI workflows and their external dependencies, emphasizing that removing malicious workflows alone may leave additional repository malware and persistence.

Article Details

Attack Vectors
  • GlassWorm reportedly used a compromised contributor account to publish four malicious OVH CLI versions outside the normal CI pipeline. Paul McCarty identified initial malicious repository pushes on March 5 but said the credential-theft date was unknown.
  • Compound payloads distribute functionality across multiple packages so individual components remain below detection thresholds. Researchers described webrunc as a possible component, not a confirmed standalone malicious package.
  • In the reported Contagious Interview activity, job-themed LinkedIn outreach leads targets to private repositories whose .npmrc redirects dependency resolution to a different registry, enabling delivery of malicious packages only in the configured repository environment.
  • GhostAction used previously planted malicious GitHub Actions workflows to add VS Code tasks that execute JavaScript disguised as a font file when developers open compromised repositories.
  • The GhostAction payload attempts to collect credential files and SSH private keys and installs a cron callback every five minutes. Researchers reported that the stealer only partially works.
  • GhostAction also reportedly phished open source maintainers using a lookalike domain.
Defensive Notes
  • Audit CI workflows, action versions, external containers, and the credentials and source code exposed to each stage; revisit copied or long-standing workflow configurations.
  • Sign commits and require signed commits where supported; the speakers cautioned that signing alone does not address malicious workflow execution.
  • Monitor releases built or published outside the established pipeline; the speakers linked the unusual OVH CLI build provenance to rapid discovery.
  • Inspect compromised repositories for added VS Code tasks and fake font payloads, and check developer systems for persistence. Removing malicious workflows alone did not remove GhostAction's additional malware.
  • Evaluate dependencies and runtime retrieval behavior together rather than relying only on isolated package signals, because compound payloads can distribute suspicious functionality across packages.
  • Investigate lingering contributor-account compromise during incident response. McCarty reported that malicious pre-install content remained in the contributor's repositories after OVH announced incident response.

MITRE ATT&CK

People

Threat Actors

Vendors

Products

Countries

Related Articles