Massive GitHub Attack: 4,800 Accounts Used to Steal Credentials
A large-scale automated campaign compromises cloud security across thousands of critical repositories and projects
November 17, 2026 · 3 min read
TL;DR: A massive campaign called 'wfx-loot' has used nearly 5,000 GitHub accounts to inject malware into over 60,000 repositories. The goal was the automated theft of cloud credentials and API keys for services like AWS, Azure, and OpenAI.
The assault on the software supply chain: A paradigm shift in cybersecurity
The modern development ecosystem, built on open collaboration and automation, is facing an unprecedented crisis of trust. According to a recent report by the security firm Aikido, the campaign known as wfx-loot has managed to compromise 61,764 GitHub repositories using 4,837 previously breached accounts. This event is not an isolated attack, but a technical demonstration of how CI/CD (Continuous Integration and Continuous Deployment) architecture has become the weakest link in the global software supply chain. Unlike historical breaches that focused on package poisoning (such as the event-stream or ua-parser-js cases), this attack uses the victim's legitimate infrastructure to self-execute credential theft, raising the sophistication of the threat vector.
The mechanics of the attack: 'wfx-loot' and automated social engineering
The modus operandi of wfx-loot relies on exploiting human and technical trust. Attackers used legitimate accounts to send pull requests under the guise of "dependency audits" (titled as docs: dependency audit notes). This social engineering tactic, executed at a massive scale via scripts, sought to identify which repositories had GitHub Actions enabled with permissive settings.
Once the repository user or maintainer accepted the PR, the malicious code executed within the GitHub Actions environment. The payload was specifically designed to search for environment variables and configuration files containing secrets. Primary targets included:
- Cloud credentials: AWS, Azure, and Google Cloud Platform tokens, which grant full control over an organization's cloud infrastructure.
- Third-party service secrets: Docker, Kubernetes, OpenAI, and Anthropic API keys, allowing attackers not only to steal data but also to incur massive operational costs through the use of external compute (cryptojacking or model training).
- Evasion tactics: After exfiltration, the attackers deleted workflow execution logs, removing any immediate forensic traces and complicating detection by security teams (SOC).
Why this is a turning point in corporate security
Historically, supply chain attacks were limited to code injection into open-source libraries (the famous 'ghost library' vulnerability). wfx-loot marks a dangerous evolution: the attacker no longer needs the user to download a malicious package; they use the company's own automation tools as a weapon. This shift turns the CI/CD system, designed to accelerate development, into a direct attack vector against the core of corporate infrastructure.
Comparatively, this incident recalls the SolarWinds breach (2020), where the software build process was compromised to distribute malware. However, wfx-loot is more insidious due to its decentralized nature and its ability to penetrate thousands of projects simultaneously, leveraging the implicit trust developers place in GitHub's automation tools.
Consequences and mitigation measures: The urgency of a 'Zero Trust' policy
Although the confirmed number of repositories where the payload was executed stands at 1,835, the real impact could be exponential due to the persistence of stolen credentials. If an attacker obtains an AWS master key, password rotation is insufficient if IAM (Identity and Access Management) roles and backdoors created in the cloud are not audited.
For organizations, mitigation requires a Zero Trust approach applied to CI/CD:
- Permission Auditing: Strictly limit GitHub Actions token permissions (using permissions: read-all by default).
- Secret Scanning: Implement tools that scan repositories in real-time before secrets are exposed in commit history.
- Behavioral Monitoring: Do not blindly trust workflow logs; integrate security solutions that alert on unusual activity in automated actions.
In conclusion, wfx-loot is not a failure of GitHub as a platform, but a failure in security management within automated workflows. Companies must understand that, in the age of AI and the cloud, unsupervised automation is essentially an open door for those who know how to ring the right doorbell.