Gheist Campaign Exploits Stolen GitHub Tokens to Exfiltrate Repository Secrets, Highlighting Supply Chain Risks

A new automated campaign, dubbed 'Gheist' by cloud security company Wiz, has been actively exploiting stolen GitHub tokens to compromise private repositories and GitHub Enterprise environments. The attackers plant malicious workflows that print repository secrets directly into GitHub Actions logs, which are then downloaded before the run is deleted. This method allows for the exfiltration of sensitive data without triggering traditional network egress alerts. The significance of this campaign lies in its direct attack on the software supply chain and its ability to bypass existing security measures. While GitHub's safeguards successfully halted public repository runs, private repositories and GitHub Enterprise Server instances were vulnerable due to a lack of similar automated protections. The compromise of a single token can grant attackers access to numerous repositories and their associated secrets, including production credentials. This incident fits into a broader trend of increasing attacks targeting development environments and supply chains. As organizations increasingly adopt DevOps and cloud-native practices, the attack surface expands to include source code repositories, CI/CD pipelines, and artifact registries. Previous incidents, such as the March compromise of the Trivy vulnerability scanner, have highlighted the vulnerability of these interconnected systems. The use of AI agents in both offensive and defensive cybersecurity is also a growing trend, with attackers leveraging AI to automate exploitation and defenders using it for vulnerability discovery and threat detection. In practice, this means that developers and security teams must re-evaluate their approach to GitHub security. Immediate actions include rotating any potentially compromised tokens and implementing more stringent access controls. Organizations should enforce mandatory review processes for environments holding production secrets and limit OIDC trust policies to protected branches. Furthermore, adopting short-lived, fine-grained tokens instead of broad personal access tokens can significantly reduce the blast radius of a compromise. The attackers' technique of exfiltrating secrets via Actions logs also necessitates enhanced monitoring of CI/CD pipeline activities and log analysis for unusual patterns. The incident serves as a stark reminder that even with robust platform-level security, the weakest link often remains credential management and the security posture of individual repositories.