Skip to main content
a promotional graphic telling you that PCI Compliance is no longer an annual exercise and that continuous monitory must be built in
4,837 Stolen GitHub Logins, 61,764 Repos ProbedIncident
5 min readFor Security Engineers

4,837 Stolen GitHub Logins, 61,764 Repos Probed

What Happened

A campaign tracked by application security company Aikido as "wfx-loot" used credentials stolen from 4,837 GitHub accounts to probe 61,764 repositories. The goal was to steal secrets and cloud credentials, including AWS keys, by injecting malicious workflows into repositories through modified GitHub Actions files. Affected repositories included those belonging to major organizations.

The attackers didn't exploit a software vulnerability. They authenticated as legitimate users and abused a feature working as designed: GitHub Actions executes whatever the workflow file specifies, with access to any secrets the workflow can read.

One detail deserves your attention. 92% of the malicious pull requests remain open. This is a detection and response failure across many repositories, long after the activity began.

Timeline

This timeline outlines the attack's logical stages, as exact dates for the initial account compromises aren't confirmed.

  1. Credential theft. Attackers obtained valid credentials for 4,837 GitHub accounts.
  2. Reconnaissance at scale. These accounts probed 61,764 repositories to identify where a workflow change would be effective.
  3. Workflow injection. Malicious workflows were introduced via modified GitHub Actions files, often delivered as pull requests.
  4. Secret exfiltration. The injected workflows were designed to harvest repository secrets and cloud credentials.
  5. Non-response. The majority of the malicious pull requests, 92%, remain open, indicating they were neither merged-and-reverted nor closed as hostile.

Which Controls Failed or Were Missing

This campaign highlights five gaps in controls.

Account authentication. Nearly 5,000 accounts were compromised. Either strong authentication wasn't enforced, or it was bypassed through stolen tokens or sessions.

Workflow change review. A modified workflow file is a code execution change. If your pipeline treats a .github/workflows/ edit like a README edit, you've got a gap.

Secret scoping. If a workflow triggered by an outside pull request can read cloud credentials, the secret is exposed to anyone who can open one.

Long-lived cloud credentials. Static AWS keys stored as repository secrets remain useful to an attacker until someone rotates them.

Detection and triage. A 92% open rate on the malicious pull requests shows that nobody was watching for or acting on hostile workflow changes.

What the Relevant Standards Require

These mappings show where your existing control framework already covers this. Check your own implementation against each.

SOC 2 Type II

  • CC6.1 requires logical access controls over protected information assets. Unenforced strong authentication on developer accounts is a direct gap.
  • CC6.6 covers protection against threats from outside your system boundary. Outside-contributor pull requests running with secret access fall here.
  • CC8.1 requires change management with authorization and review. Unreviewed workflow modifications fail it.
  • CC7.2 and CC7.3 address monitoring for anomalies and evaluating security events. Hostile pull requests sitting open for weeks indicate neither is working for this surface.

ISO/IEC 27001:2022 (Annex A)

  • A.5.17 covers authentication information, which includes the secrets and tokens at the center of this attack.
  • A.8.2 and A.8.5 cover privileged access rights and secure authentication.
  • A.8.4 covers access to source code, which includes who can alter pipeline definitions.
  • A.8.32 covers change management.
  • A.8.15 and A.8.16 cover logging and monitoring activities.

NIST Cybersecurity Framework (CSF) v2.0

  • PR.AA (Identity Management, Authentication, and Access Control) covers authenticating and authorizing users and services.
  • DE.CM (Continuous Monitoring) covers watching assets for adverse events. Your repositories are assets.
  • DE.AE (Adverse Event Analysis) is the function that should have flagged the hostile pull requests.
  • PR.PS covers platform security, including configuration of the tools in your pipeline.

PCI DSS v4.0.1, if your repositories touch cardholder data environments:

  • Requirement 8.4.2 requires multi-factor authentication for all access into the cardholder data environment.
  • Requirement 6.5.1 requires change control procedures for system components, including review and approval.
  • Requirement 8.6.2 prohibits hard-coded passwords in scripts, configuration files, and source code. Secrets sitting in workflow files or repository settings with broad reach run against this.
  • Requirement 10.4.1 requires log reviews, and 10.7.2 requires failures of critical security controls to be detected and responded to.

Lessons and Action Items for Your Team

1. Enforce phishing-resistant MFA on every GitHub account. Prefer hardware keys or passkeys over SMS or one-time codes. Require it at the organization level, and audit for members who aren't compliant.

2. Treat workflow files as high-risk code. Add a CODEOWNERS entry covering .github/workflows/ and require review from your security or platform team. Pair this with branch protection so no one, including admins, can merge workflow changes unreviewed.

3. Lock down workflow permissions. Set the default GITHUB_TOKEN to read-only at the organization level, and grant write scopes per job only where required. Restrict which actions are allowed to run, and pin third-party actions to full commit SHAs rather than mutable tags.

4. Keep secrets away from untrusted triggers. Don't expose secrets to workflows triggered by pull requests from forks or outside contributors. Be especially careful with pull_request_target, which runs with elevated context. Require manual approval before first-time contributor workflows run.

5. Replace static cloud keys with short-lived credentials. For AWS, use GitHub's OIDC integration with IAM role trust policies scoped to specific repositories and branches. A stolen workflow then yields a token that expires, not a key that works for months. Where static keys remain, rotate them on a schedule and immediately after any suspected exposure.

6. Detect workflow tampering. Alert on creation or modification of any file under .github/workflows/, on new outbound network destinations from runners, and on pull requests that touch workflows from accounts with no prior history in the repository. Stream GitHub audit logs to your SIEM.

7. Triage open pull requests like an alert queue. If 92% of the hostile pull requests are still open, nobody owns that queue. Assign ownership, set a review window, and report on stale or suspicious pull requests as a metric.

8. Check your own exposure now. Search your organization for workflow changes you didn't authorize, review recent pull requests that touch pipeline files, and examine cloud audit logs for credential use from unfamiliar sources. If you find a malicious workflow, assume any secret it could read is compromised and rotate it.

Your pipeline holds the keys to your cloud environment, and the standards you already audit against cover that. Apply the same access, change-control, and monitoring requirements to your workflow files that you apply to production systems, because that's what they are.

Topics:Incident
Digital advertisement promoting the whitepaper “The State of Application Security in Modern Software,” showing the cover f the whitepaper and text highlighting AppSec risks, AI code threats, API vulnerabilities, and a button to download the whitepaper.

You Might Also Like