Wfx-loot campaign used 4,837 compromised GitHub accounts to steal cloud credentials

· Original article ↗

Summary

Aikido researchers say wfx-loot used credentials from 4,837 compromised GitHub accounts to probe 61,764 repositories and steal secrets from at least 1,835 targets through malicious GitHub Actions workflows.

Key points

  • Aikido says the campaign began on November 10, 2026, and used compromised accounts to probe repositories for write access through disguised pull requests.
  • Researchers report that the actor modified or added GitHub Actions workflows, or planted an npm install hook, to deliver a credential-stealing script.
  • The script targeted GitHub Actions tokens, cloud credentials, developer and infrastructure secrets, AI service credentials, and secrets in files and runtime data.
  • The script could exchange GitHub OIDC tokens for temporary AWS, Azure, and Google Cloud sessions; AWS role chaining and GCP identity impersonation could support further access where permissions allowed.
  • The actor compressed collected data into tarballs and exfiltrated it through HTTP POST, then reverted changes and deleted workflow runs and logs; Aikido says cleanup took under a minute.
  • Automated publishing carried payloads into Packagist’s digitalelvis/neuronai-studio 3.14.3 and Go module github[.]com/onuragtas/docker-env versions v1.0.278 and v1.0.280.
  • Aikido advises repository owners to check for the wfx-docs branch, probe pull requests, and gaps in workflow run numbers; where execution is confirmed, available secrets and cloud roles should be treated as compromised and rotated.

Article Details

Attack Vectors
  • The attacker used stolen credentials from 4,837 compromised GitHub accounts to probe repositories and open pull requests where those accounts had write access.
  • Malicious code was introduced by modifying existing GitHub Actions workflows, adding workflows on short-lived branches, or planting an npm install hook.
  • A shell script collected GitHub Actions secrets, cloud credentials, developer credentials, and other secrets, including from environment variables, files, private keys, shell histories, process memory, and workflow artifacts.
  • The script exchanged GitHub OIDC tokens for temporary cloud sessions, chained into AWS IAM roles, and used collected service-account credentials to impersonate identities in GCP.
  • Collected data was compressed into tarballs and exfiltrated via HTTP POST to attacker-controlled endpoints.
  • The attacker reverted changes and deleted workflow runs and logs after payload execution.
Defensive Notes
  • Repository owners should look for the ‘wfx-docs’ branch, the probe pull request, and gaps in workflow run numbers.
  • Where payload execution is confirmed, treat all secrets and cloud roles available to affected workflows as compromised and rotate them immediately.

MITRE ATT&CK

Products

Countries

Industries

Related Articles