Applying zero trust principles to self-hosted build agents in cloud environments

Sammanfattning

Unlike platform-hosted build agents, self-hosted CI/CD build agents are managed by the organization that uses them. Self-hosted CI/CD build agents give organizations control over the build environment, but they also create security risks because they can connect source code, workflow execution, administrative access, credentials, and cloud resources. This thesis investigates how Zero Trust principles can reduce threats to self-hosted build agents in cloud environments. The study uses a before-and-after comparative case study in a controlled GitHub Actions and Microsoft Azure environment. A baseline self-hosted runner setup was analyzed through threat modeling using STRIDE, and the identified weaknesses were used to design an improved environment based on Zero Trust principles, including stronger verification, more restricted access, and measures that limit the effect of compromise. The evaluation compared the baseline and improved conditions using repeated attack scenarios and exposure scoring across attack feasibility, access level, and affected assets or evidence. The results show that the baseline runner was exposed to host compromise, credential reuse, persistent workflow state, broad Azure permissions, and limited visibility. The improved setup reduced exposure in all evaluated scenarios through controls such as removing direct public administrative exposure, strengthening identity-based access, reducing residual runner state, replacing reusable cloud secrets with federated authentication, narrowing cloud permissions, and improving centralized logging. The thesis shows that Zero Trust principles can provide a practical structure for reducing the attack surface and limiting the impact of compromise in self-hosted CI/CD environments.