GitHub credential theft campaign plants stealer in CI workflows
Attackers used 4,837 stolen GitHub accounts to test write access on over 61,000 repositories, planting a secret-stealing script in at least 1,806 of them to harvest cloud credentials and tokens from CI/CD pipelines.

A large-scale attack has hijacked thousands of GitHub accounts to plant a secret stealer in continuous integration workflows. Researchers at Aikido Security and SafeDep uncovered a campaign that used 4,837 stolen credentials to test write access across 61,764 repositories starting Tuesday. The attackers pushed a malicious script into at least 1,806 repositories, aiming to harvest cloud credentials and access tokens.
Each attack began with a pull request titled docs: dependency audit notes, opened from a branch named wfx-docs. This was a probe to verify the hijacked account had write permissions. In a separate analysis, SafeDep counted 8,093 repositories that received malicious commits between 12:46 UTC on Tuesday and 04:36 UTC today. The stealer followed in repositories where the test succeeded.
Stealer capabilities and reach
The malicious script targeted the credentials accessible to continuous integration workflows. It collected cloud roles, tokens, and secrets from CI runners. Aikido Security confirmed the stealer executed in at least 1,806 repositories, meaning credentials were likely exfiltrated. The attackers employed two primary methods to trigger the script.
| Deployment Method | Number of Repositories |
|---|---|
| Added to existing workflow | 7,309 |
| npm postinstall hook | 771 |
The npm postinstall hook is particularly concerning as it runs on a developer's local machine during an npm install command, extending the attack surface beyond the CI environment.
Indicators of compromise
Repository maintainers must check for specific artifacts linked to this campaign. Activity from Tuesday onward should be scrutinized for these signs.
| Indicator Type | Specific Examples |
|---|---|
| Branch names | wfx-docs, chore/[a-z0-9]{8}, ci-check-[0-9]{4} |
| Pull request title | docs: dependency audit notes |
| Commit messages | ci: refresh build cache, chore: add ci helper, chore: add dependency audit workflow, chore: initial commit, ci: restore, ci: cleanup |
| Script SHA-256 | 89695abb392d5cfd4972672961e644bb46baa98aa1ed7173c2a2816c762735cf |
A more reliable sign of execution is a missing run number on the attacker's commit. This can signal the stealer ran even if teams find no wfx-loot artifacts or related cache entries.
Exfiltration and persistence
The stolen data was sent to attacker-controlled infrastructure. SafeDep researchers found the stealer sends data via HTTP POST to specified IP addresses or to a Telegram chat. In a sample of 200 payload commits, about one-fifth used Telegram only. The malware also copies the entire repository, its last 100 commits, and recent workflow artifacts.
The stealer was designed for persistence. It contains code that can exchange GitHub OpenID Connect tokens for active sessions in AWS, Google Cloud, and Azure. This mechanism could allow attackers to retain access even after teams rotate static credentials. Aikido Security noted, "we don't attribute November to the same actor at this stage," suggesting this may be a distinct operation.
Recommended response and mitigation
Owners of potentially affected repositories must take immediate action. SafeDep's researchers advise revoking all tokens, OAuth app grants, SSH keys, and sessions associated with the commit author, and enabling two-factor authentication. Every secret and environment variable the repository could read must be rotated. If the stealer executed, treat every secret and cloud role available to that workflow as fully compromised.
Organizations should audit collaborator and organization member access. Aikido Security found 10,431 of the test pull requests came through this path, and 92% of those were still open when they checked. Teams must also inspect their cloud environments for specific signs of compromise.
These include Entra ID federated credentials named wfx-ci-, AWS CloudTrail events for AssumeRoleWithWebIdentity or AssumeRole sessions named wfx- followed by CreateAccessKey calls, and Google Cloud GenerateAccessToken operations from a GitHub workload identity pool. SafeDep also identified version 3.14.3 of the Packagist package digitalelvis/neuronai-studio as malicious. While a standard composer install does not trigger the stealer, running npm install inside the package's vendor directory does. Users should block version 3.14.3, pin to version 3.14.2 or earlier, and treat any machine that installed or opened the affected version as compromised.





