Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

IncidentSupply chain

wfx-loot: stolen GitHub accounts pushed a CI secret stealer into tens of thousands of repositories, taking cloud keys and AI tool tokens

CriticalAikido SecurityGitHub

Aikido Security reported on November 12, 2026 that since November 10 an attacker with stolen GitHub credentials had used 4,837 accounts to open probe pull requests titled "docs: dependency audit notes", from a wfx-docs branch, in 61,764 repositories; 10,431 went into repositories reached through collaborator or organization access, and 92 percent were still open. Where the probe confirmed write access, the attacker added a shell script of about 1,500 lines to an existing workflow (commit "ci: refresh build cache"), as an npm postinstall hook ("chore: add ci helper") or in a new workflow on a throwaway branch ("chore: add dependency audit workflow"), then reverted it within about a minute and deleted the run and its logs. Aikido counts at least 1,835 repositories where the payload ran. It collects Actions secrets and variables, OIDC tokens, cloud, Kubernetes, registry, SSH and Vault credentials, shell histories and tokens for AI tools including OpenAI, Anthropic, Gemini, Hugging Face, Cursor, Copilot, Codex and Ollama, and sends them over HTTP to two servers. It mints cloud sessions, creates AWS access keys, posts forged success checks to get past branch protection, sets an AWS role's trust policy to accept any AWS account and adds federated identity credentials named wfx-ci to Azure applications. At least 50 affected users publish packages; the Packagist package digitalelvis/neuronai-studio 3.14.3 carried the payload, and two docker-env Go tags carry it only in a workflow file that Go builds never run.

Replay the attack step by step in the Learning Center

How it happened

  1. Stolen GitHub credentials opened a harmless-looking pull request to find every repository the account could write to.
  2. A short-lived workflow change or install hook ran a secret stealer in CI, then the commit was reverted and the run and logs deleted.
  3. Stolen cloud tokens were turned into lasting access: new AWS keys, an AWS role open to any account and new Azure federated credentials.

Impact

Secrets, cloud roles and AI tool tokens from at least 1,835 repositories, with AWS and Azure changes that survive rotating the secrets.

The fix

Aikido advises checking every repository you can write to for the probe pull request, a wfx-docs branch and the commit titles above from November 10 on, and a missing run number on the attacker's commit. Where the payload ran, treat every secret and cloud role its workflows could reach as compromised, rotate them including AI provider keys, remove AWS trust policy changes, IAM access keys created by the attacker and wfx-ci federated credentials in Azure, and treat any machine that installed neuronai-studio 3.14.3 as compromised.

Lessons

  • A CI runner holds cloud and AI keys; give each workflow only the secrets and roles it needs.
  • Deleted runs and reverted commits hide the attack; look for gaps in run numbers.
  • Rotating secrets does not undo trust policies and federated credentials an attacker added.

Sources