A sophisticated credential-stealing campaign has compromised tens of thousands of GitHub repositories since November 10, 2026. Security researchers at Aikido identified the operation, dubbed wfx-loot, which leverages stolen GitHub credentials to inject malicious payloads into automated workflows. The attacker’s primary objective is to harvest GitHub Actions secrets, cloud infrastructure credentials, and tokens for various AI development tools, including OpenAI, Anthropic, Cursor, and Copilot.
The campaign is characterized by its high volume and rapid execution. Researchers observed 4,837 compromised accounts initiating probe pull requests across 61,764 repositories. Of these, over 10,000 pull requests targeted repositories where the compromised account held collaborator or organization access, demonstrating how a single stolen token can be used to infiltrate numerous projects. In more than 1,800 instances, the malicious script successfully executed, exfiltrating sensitive data directly to attacker-controlled servers via HTTP.
The attacker follows a consistent methodology to maintain stealth. The process typically begins with a pull request titled "docs: dependency audit notes" to verify that the stolen token possesses write access. Once confirmed, the attacker injects a 1,500-line shell script into existing workflows, hooks it into npm install processes, or creates new workflows on temporary branches. To minimize detection, the attacker reverts these changes, deletes workflow runs, and clears logs within one minute, often leaving behind only a skipped sequence number in the workflow history.
Beyond simple credential theft, the malware performs aggressive persistence maneuvers within cloud environments. It creates long-lived AWS access keys, modifies AWS role trust policies to allow access from any account, and adds federated identity credentials to Azure applications. These modifications are particularly dangerous because they remain active even after standard secret rotation procedures are performed, requiring manual intervention to identify and remove the unauthorized access points in cloud configurations.
The campaign has also impacted the software supply chain by injecting payloads into published packages. Affected projects include version 3.14.3 of the Packagist package digitalelvis/neuronai-studio and versions v1.0.278 and v1.0.280 of the Go module github.com/onuragtas/docker-env. While the malicious code in the Go module was confined to a workflow file—meaning it would not be executed during standard package builds—researchers warn that any machine that installed or opened the compromised Packagist package should be considered fully compromised.
There is evidence of overlap with previous security incidents, specifically the GhostAction campaign from October. Researchers noted that accounts previously targeted by GhostAction were significantly more likely to be hit by wfx-loot, with the new attacker even injecting their payload into the workflow files left behind by the earlier campaign. Despite this overlap, current analysis suggests that wfx-loot and GhostAction are likely the work of distinct threat actors.
To determine if a repository has been compromised, developers should audit their environments for the existence of a "wfx-docs" branch or pull requests and commits with titles such as "ci: refresh build cache," "chore: add ci helper," or "chore: add dependency audit workflow." Because the attacker deletes artifacts quickly, the absence of these indicators does not guarantee safety.
The affected party has not yet issued a comprehensive public statement regarding the scope of the breach, but the nature of the attack confirms that the threat is widespread. Security experts advise that if any evidence of the payload is found, all secrets and cloud roles associated with the affected workflows must be treated as compromised. Given the attacker's ability to manipulate cloud trust policies and federated identities, remediation requires a thorough audit of cloud infrastructure beyond simply rotating GitHub secrets.





