- by x32x01 ||
wfx-loot: 4,837 GitHub Accounts Compromised in a Massive AI and Cloud Credential Theft Campaign
A massive supply chain attack called wfx-loot has targeted thousands of GitHub accounts, putting tens of thousands of software repositories at risk. Instead of exploiting a vulnerability in GitHub itself, attackers used stolen developer credentials to access repositories, inject malicious code into GitHub Actions workflows, and steal sensitive credentials.
According to security researchers at Aikido Security, the campaign involved:
The Pull Requests used a specific title:
They also came from a branch named:
At first glance, these changes looked like routine documentation or dependency audit updates.
However, their real purpose was to identify repositories where the compromised accounts had permission to write code.
Once attackers found a repository with sufficient permissions, they moved to the next stage:
Its targets included:
The malware also attempted to access credentials from running processes and cloud authentication mechanisms.
This is especially concerning for developers who connect AI coding agents to repositories, API services, and cloud infrastructure.
A compromised API key can expose more than an AI account if it provides access to additional services or resources.
Researchers identified several techniques, including:
This creates an important security lesson: removing malicious code does not necessarily remove the attacker's access.
Even rotating API keys may not be enough if the attacker has already created additional access methods.
Instead:
AI tools such as Claude, Cursor, GitHub Copilot, and Codex can interact with repositories, development environments, and external services.
When these tools receive access to API keys, GitHub tokens, or cloud credentials, a compromised environment can expose much more than source code.
Developers should:
It also highlights a growing challenge as AI coding agents gain access to more development tools and infrastructure.
The real security question is no longer just whether your GitHub account is protected.
It's whether you know exactly what an attacker could access if that account, one of your API keys, or an AI agent were compromised.
A massive supply chain attack called wfx-loot has targeted thousands of GitHub accounts, putting tens of thousands of software repositories at risk. Instead of exploiting a vulnerability in GitHub itself, attackers used stolen developer credentials to access repositories, inject malicious code into GitHub Actions workflows, and steal sensitive credentials.
According to security researchers at Aikido Security, the campaign involved:
- 4,837 compromised GitHub accounts.
- 61,764 repositories targeted with malicious Pull Requests.
- At least 1,800 repositories where researchers identified malware execution and credential theft.
🔍 How the wfx-loot GitHub Attack Worked
The attack began with compromised GitHub accounts. Attackers used these accounts to send seemingly harmless Pull Requests to repositories they could access.The Pull Requests used a specific title:
docs: dependency audit notesThey also came from a branch named:
wfx-docsAt first glance, these changes looked like routine documentation or dependency audit updates.
However, their real purpose was to identify repositories where the compromised accounts had permission to write code.
Once attackers found a repository with sufficient permissions, they moved to the next stage:
- Injected malicious code into GitHub Actions workflows, an npm installation hook, or a temporary workflow.
- Executed a shell-based credential stealer inside the affected environment.
- Collected secrets and credentials accessible to the CI/CD runner.
- Attempted to expand access to connected cloud environments.
- Reverted malicious changes and deleted workflow execution logs to hide evidence.
🔑 What Information Did the Malware Steal?
The malware was designed to collect much more than GitHub passwords.Its targets included:
| Category | Targeted credentials |
|---|---|
| AI services | OpenAI, Anthropic, Gemini, Hugging Face |
| AI development tools | Claude, Cursor, GitHub Copilot, Codex |
| Cloud platforms | AWS, Microsoft Azure, Google Cloud |
| GitHub | Personal Access Tokens, Actions secrets, workflow tokens |
| Developer environments | SSH keys, .env files, configuration secrets |
| CI/CD systems | GitHub Actions secrets, variables, and runtime tokens |
This is especially concerning for developers who connect AI coding agents to repositories, API services, and cloud infrastructure.
A compromised API key can expose more than an AI account if it provides access to additional services or resources.
⚠️ Why This Attack Is More Dangerous Than It Looks
The most concerning part of wfx-loot is that attackers did not stop after stealing credentials. They also attempted to establish persistent access to cloud environments.Researchers identified several techniques, including:
- AWS: Creating long-lived access keys and modifying IAM role trust policies.
- Microsoft Azure: Adding federated identity credentials that could allow future authentication without a stored secret.
- Cloud access: Using stolen tokens to obtain sessions and access additional resources.
This creates an important security lesson: removing malicious code does not necessarily remove the attacker's access.
Even rotating API keys may not be enough if the attacker has already created additional access methods.
🛡️ How to Protect Your GitHub Account and Repositories
Developers and repository maintainers should take the following precautions:1. Secure your GitHub account
- Enable two-factor authentication (2FA) or Passkeys.
- Use a password manager and unique passwords.
- Review active sessions, personal access tokens, SSH keys, and authorized OAuth applications.
- Revoke credentials you do not recognize.
2. Inspect suspicious Pull Requests
Check repositories you own or have permission to modify for:- Branches named
wfx-docs. - Pull Requests titled
docs: dependency audit notes. - Unexpected commits titled
ci: refresh build cacheorchore: add ci helper. - Unexplained gaps in GitHub Actions workflow history.
3. Restrict GitHub Actions permissions
Use the principle of least privilege:- Give workflows only the permissions they actually need.
- Avoid unnecessary write access through
GITHUB_TOKEN. - Review third-party Actions before using them.
- Pin Actions to trusted commit SHAs when possible.
- Avoid exposing sensitive secrets to untrusted Pull Request workflows.
4. Protect API keys and cloud credentials
Use short-lived credentials whenever possible, limit their permissions, and avoid sharing production secrets across unrelated workflows.5. Respond properly if malware execution is confirmed
Do not simply delete suspicious files.Instead:
- Disable affected workflows and investigate the compromised repository.
- Revoke and rotate every exposed API key, token, and cloud credential.
- Review AWS IAM policies, access keys, and role trust relationships.
- Inspect Azure federated identity credentials for unauthorized additions.
- Review cloud audit logs for suspicious authentication and privilege changes.
- Rebuild affected environments from trusted sources when necessary.
🤖 What This Means for AI Developers
This campaign highlights an important security risk for developers building AI-powered applications and autonomous coding agents.AI tools such as Claude, Cursor, GitHub Copilot, and Codex can interact with repositories, development environments, and external services.
When these tools receive access to API keys, GitHub tokens, or cloud credentials, a compromised environment can expose much more than source code.
Developers should:
- Keep AI API keys separate from production infrastructure credentials.
- Avoid giving AI agents unrestricted cloud permissions.
- Use isolated development environments.
- Never expose secrets unnecessarily to AI tools or untrusted workflows.
- Audit which repositories and external services each agent can access.
❓ Frequently Asked Questions
----------------------Is wfx-loot exploiting a GitHub vulnerability?
No. The campaign relies on stolen GitHub credentials and the permissions associated with compromised accounts. Researchers have not identified this as an underlying GitHub platform vulnerability.Does closing a suspicious Pull Request remove the threat?
No. Closing the Pull Request does not revoke stolen credentials or remove malicious code that may already have executed in a workflow.Can rotating API keys completely stop the attack?
Not necessarily. Attackers may have created persistent cloud access through new AWS access keys or Azure federated credentials. These must also be investigated and removed.Are AI API keys specifically targeted?
Yes. Researchers reported targeting credentials associated with OpenAI, Anthropic, Gemini, and other AI development services.🔎 Final Thoughts
The wfx-loot campaign demonstrates how stolen developer credentials can become an entry point into thousands of software repositories and connected cloud environments.It also highlights a growing challenge as AI coding agents gain access to more development tools and infrastructure.
The real security question is no longer just whether your GitHub account is protected.
It's whether you know exactly what an attacker could access if that account, one of your API keys, or an AI agent were compromised.