26,400 Valid Cloud and AI Credentials Exposed on Gitee

Introduction

An independent security researcher has uncovered a massive exposure of valid authentication credentials sitting in plain sight across public code repositories on Gitee, China's largest code-hosting platform. The finding reveals that basic secrets hygiene remains a critical blind spot for development organizations worldwide, with particular implications for cloud infrastructure and artificial intelligence workloads.

The Discovery and Scope

On October 28, researcher Ben Zimmermann reported the results of a comprehensive scan covering 7.25 million public Gitee repositories. The scan identified 26,400 credentials that were not only present in source code but remained fully functional when tested. This means any actor who accessed the public repositories could have immediately used these credentials to authenticate against the corresponding services.

The scale of the exposure is significant. Gitee hosts millions of repositories from individual developers, enterprises, and open-source projects. Unlike GitHub, which has invested heavily in automated secret scanning and push protection, Gitee's public repository security tooling has historically been less visible to the international security community. Zimmermann's research brings much-needed attention to this gap.

Types of Credentials Exposed

The majority of the valid credentials discovered were keys for cloud and AI services. This alignment is not coincidental. Modern development workflows increasingly rely on programmatic access to cloud provider APIs, managed AI model endpoints, and machine learning platform credentials. Developers working rapidly to integrate these services often hardcode API keys, access tokens, and service account credentials directly into application code during testing phases, then forget to remove them before pushing to remote repositories.

Cloud credentials typically grant access to compute resources, storage buckets, databases, and identity management systems. AI service credentials often provide access to model inference endpoints, training pipelines, and proprietary datasets. Both categories represent high-value targets for threat actors seeking to hijack compute resources for cryptocurrency mining, exfiltrate sensitive data, or poison machine learning models.

Risk Assessment and Impact

The immediate risk is direct unauthorized access. Valid credentials in public repositories are equivalent to leaving house keys under a doormat with the address posted online. Threat actors routinely scan public code hosting platforms for exposed secrets using automated tooling. The 26,400 validated credentials represent confirmed compromise opportunities—there is no theoretical barrier to exploitation.

Beyond direct access, exposed credentials enable supply chain attacks. A compromised cloud service account can be used to inject malicious code into CI/CD pipelines, modify infrastructure-as-code templates, or push tainted container images. For AI workloads, exposed credentials could allow adversaries to manipulate training data, steal proprietary models, or deploy backdoored models to production environments.

The regulatory implications are also substantial. Organizations subject to data protection regulations such as GDPR, China's Personal Information Protection Law, or industry-specific frameworks may face compliance violations if exposed credentials lead to data breaches. The fact that credentials were exposed in public repositories on a Chinese platform adds jurisdictional complexity for multinational organizations.

Practical Mitigation Steps

Organizations must treat this discovery as a catalyst for immediate action across three fronts:

• Immediate rotation and revocation: Any credential that has ever been committed to a public repository must be treated as compromised. Rotate all such credentials immediately and audit access logs for anomalous activity during the exposure window.
• Automated secret scanning: Implement pre-commit hooks, CI/CD pipeline scanning, and repository-level secret detection. Tools such as GitLeaks, TruffleHog, and commercial solutions should be mandatory gates in every development workflow.
• Developer education and guardrails: Provide secure credential management patterns—environment variable injection, secret managers (HashiCorp Vault, AWS Secrets Manager, Azure Key Vault), and short-lived token workflows. Make the secure path the easiest path.
• Repository hygiene audits: Conduct historical scans of all public and private repositories. Git history retains deleted credentials; rewriting history or rotating exposed secrets is essential even for past commits.
• Platform-level controls: Enable push protection where available. For Gitee specifically, organizations should evaluate whether the platform's native secret scanning capabilities meet their requirements or whether additional tooling is needed.

Conclusion

The exposure of 26,400 valid credentials on Gitee is a stark reminder that secrets sprawl remains one of the most exploitable and under-addressed risks in software development. As cloud and AI adoption accelerates, the volume and sensitivity of programmatic credentials will only increase. Organizations must move beyond reactive scanning to proactive, architecture-level controls that make credential exposure structurally difficult. Grivyonx helps engineering teams embed these controls directly into DevSecOps pipelines and cloud-native platforms, turning secrets management from a recurring fire drill into a solved engineering problem.