4,837 Hijacked GitHub Accounts Ran a CI Secret Stealer in 1,806 Repos: How SMBs Can Detect and Block wfx-loot

4,837 Hijacked GitHub Accounts Ran a CI Secret Stealer in 1,806 Repos: How SMBs Can Detect and Block wfx-loot

On November 10, 2026, an attacker started opening pull requests titled “docs: dependency audit notes” from a branch named wfx-docs across GitHub. Within 26 hours, 4,837 hijacked accounts had probed 61,764 repositories, pushed malicious commits into more than 8,000 of them, and exfiltrated secrets from at least 1,806 — including repos of SMBs that never thought they were “big enough” to be a target. The campaign, now tracked as wfx-loot, is a reminder that your CI pipeline is only as strong as the token that runs it. Roughly 58,900 probe PRs are still sitting open right now, which means most repository owners have never looked.

What Happened: The Probe PR That Exposed Write Access

The first stage was almost boring: a pull request that adds a harmless docs/wfx-audit.md file. Its only real purpose was to answer one question — does this stolen token still have write access? Because the accounts were real employee and contractor accounts (many with collaborator or org-level rights), the probe also fired any pull_request workflows you had configured. Researchers at Aikido Security named the campaign wfx-loot; Endor Labs relayed one victim report tracing the initial access to a leaked GitHub CLI OAuth token, and StepSecurity observed probe PRs attributed to the Fly.io GitHub App. The lesson: the entry point was credentials, not a zero-day. The first probes hit at 12:45 UTC on November 10, and by the afternoon of November 11 the campaign had touched 61,764 repositories — including 10,431 PRs aimed at repos the compromised accounts did not own outright, reached only through collaborator and organization permissions. That is the part that should worry lean teams: you do not need to be a customer of the attacker, you just need to have given a contractor write access once.

The Kill Chain: Probe, Inject, Exfiltrate, Erase

  1. Probe. A wfx-docs branch plus a PR titled docs: dependency audit notes confirms write access and maps the repo’s workflow surface.
  2. Inject. Commits titled ci: refresh build cache, chore: add ci helper, chore: add dependency audit workflow, ci: restore, or ci: cleanup push a secret-stealing workflow and a hidden .ci-*.sh helper, sometimes wired through postinstall in package.json.
  3. Exfiltrate. The workflow dumps environment secrets, registry tokens, and cloud credentials from the runner.
  4. Erase. The payload cleans traces, so gaps in your workflow run numbers are often the only signal left.

This is the same class of risk we covered in our guide to supply chain vulnerabilities in SMB CI/CD pipelines — but here the “third party” is your own compromised identity. At least 50 hijacked accounts published to npm, PyPI, Go, and crates.io, so downstream installs were also exposed.

Detection: Five Commands to Run Right Now

# 1. Probe PRs in your organization
gh search prs "docs: dependency audit notes" --owner <org> --created ">=2026-11-10"

# 2. Attacker commits in a fresh clone (all branches)
git log --all --since=2026-11-10 --format='%h %ad %an %s' \
  | grep -E 'ci: refresh build cache|chore: add ci helper|chore: add dependency audit workflow|ci: restore|ci: cleanup'

# 3. Leftover attacker branches
git ls-remote origin 'refs/heads/wfx-docs' 'refs/heads/chore/*' 'refs/heads/ci-check-*'

# 4. Hidden CI helpers wired into package.json
grep -rn --include=package.json -E '"postinstall"\s*:\s*".*\.ci-[a-z0-9]{6}\.sh"' .

# 5. Workflow runs you never triggered (audit gap)
gh run list --limit 200

Cross-check the run numbers against your merge history. A workflow run with no matching PR or push from your team is a red flag. If you sign artifacts, verify your signing keys were not accessed; our SBOM and image signing walkthrough shows where those keys live and how to rotate them.

Hardening: Make Your CI Survive a Stolen Token

Start with a fail-closed policy that blocks the probe pattern itself, so a compromised contributor account cannot even open the canary PR:

name: block-wfx-probe
on: pull_request
jobs:
  deny:
    runs-on: ubuntu-latest
    steps:
      - name: Fail known wfx-loot probe patterns
        run: |
          if [[ "${{ github.head_ref }}" == wfx-docs* ]] || \
             [[ "${{ github.event.pull_request.title }}" == "docs: dependency audit notes" ]]; then
            echo "::error::Probe PR pattern matched (wfx-loot campaign)"
            exit 1
          fi
  • Require reviews for every workflow change. CODEOWNERS must own .github/workflows/; no self-merge, ever.
  • Shorten token life. Use OIDC federation instead of long-lived PATs, and store what remains in a secrets manager, per our secrets management guide.
  • Turn on push protection and secret scanning at the org level, and pin GitHub Actions to full SHAs.
  • Cap blast radius. Give jobs the minimum cloud scope; a build runner never needs production database credentials.
  • Audit collaborators monthly. Stale contractor tokens are exactly what the attacker used at scale.

If You’re Already Hit: Rotate in This Order

  1. Revoke the GitHub token family (and any OAuth app it authorized).
  2. Rotate every secret the workflow could see: registry, cloud, deploy keys.
  3. Delete attacker branches and force-push clean history; re-run builds from known-good commits.
  4. Audit downstream: dependency publishes, releases, and any artifact signed after the first probe PR.
  5. Verify in the audit log that no new PATs, OAuth grants, or deploy keys were created after the first probe, then re-check the previous week’s workflow runs for anything you do not recognize.

Want a second pair of eyes on your pipeline before the next campaign lands? Book a free 30-minute DevOps consultation and we’ll review your repo permissions, workflow gating, and secret hygiene together.

Scroll to Top