Executive brief
FastGPT, an AI application platform, contains a security flaw in its automated development pipeline. The system incorrectly allows code from untrusted external contributors to be automatically built and then deployed using high-privilege credentials. This could allow an attacker to inject malicious code into the platform's official container registry or deploy unauthorized versions of the software to the company's hosting environment, potentially compromising infrastructure and data integrity.
Technical details
The vulnerability exists in the GitHub Actions configuration, specifically within the interaction between low-privilege build workflows and high-privilege `workflow_run` jobs. Workflows such as `preview-docs-build.yml` and `preview-fastgpt-build.yml` build Docker images from untrusted pull request code. Subsequently, privileged workflows (`preview-docs-push.yml` and `preview-fastgpt-push.yml`) download these artifacts and use `GITHUB_TOKEN` and `KUBE_CONFIG_CN` secrets to push images to GHCR and deploy them to Kubernetes. Because there is no trusted-origin gate or maintainer approval required between the build and the privileged deployment, an attacker can supply a malicious Docker build context that is then executed/deployed with the repository's authority.
Affected products
- labring FastGPT <= 22ebfacbb43311e9b73294040ae0eb87390c6bba
Timeline
- 2026-05-26: advisory: Original GitHub Security Advisory published
- 2026-07-15: disclosed: CVE-2026-50562 published to NVD