CI/CD Abuse Detector: New Open-Source Tool Uses LLM to Spot Stolen Credential Attacks on Pipelines
CI/CD Abuse Detector, an open-source tool using a large language model, flags suspicious changes to CI/CD pipelines to combat supply-chain attacks that exploit stolen developer credentials.
A new open-source tool called CI/CD Abuse Detector uses a large language model to identify suspicious modifications to continuous integration and continuous deployment pipelines, workflows, and automation configurations. The project, available on GitHub, provides drop-in templates for GitHub Actions, GitLab CI, and Azure DevOps, aiming to stop a common attack vector in software supply chain compromises.
When attackers steal developer credentials, they can push unauthorized changes to workflow files. These malicious modifications often exfiltrate secrets stored in the CI environment, such as API keys, cloud provider tokens, or database passwords. CI/CD Abuse Detector analyzes the content of proposed changes using an LLM to determine whether they deviate from normal patterns or introduce suspicious commands, like curl or wget calls to unknown endpoints.
The tool is designed to be integrated directly into the CI/CD pipeline itself. It runs as a step in the workflow, examining each pull request or commit that alters workflow definitions, automation scripts, or configuration files. If the LLM detects anomalous activity, the build can be flagged, blocked, or sent for manual review, providing a proactive defense before secrets are stolen.
Stealing CI/CD secrets via modified workflow files has become a hallmark of sophisticated supply-chain attacks. In 2024, the compromise of the tj-actions/changed-files GitHub Action demonstrated how malicious pull requests targeting popular actions could expose credentials across hundreds of organizations. CI/CD Abuse Detector specifically addresses this pattern by monitoring changes to the pipeline configuration itself—the very file that dictates what the CI system executes.
The open-source nature of the tool allows security teams and platform engineers to customize detection rules, add bespoke patterns, or fine-tune the LLM for their specific environment. The project's repository encourages community contributions and includes documentation for easy deployment across major CI platforms. By lowering the barrier to implementing runtime detection of pipeline abuse, CI/CD Abuse Detector helps organizations protect one of their most sensitive attack surfaces.
While not a silver bullet against all supply chain threats—it won't stop dependency confusion or direct repository compromises—it fills a critical gap in the defense stack. Human review of every pipeline change is impractical at scale, and simple regex-based checks are brittle. LLM-based analysis can understand context, reducing false positives while catching novel attack patterns. As credential theft remains persistent, tools like this shift the advantage back to defenders by automating detection of known supply chain abuse techniques.