26,400 Valid Cloud and AI Credentials Exposed in Public Gitee Repositories

An independent security researcher has uncovered a significant cache of 26,400 active cloud and AI service credentials that were inadvertently exposed in public repositories hosted on Gitee, a popular Chinese code-sharing platform. The credentials, identified by researcher Ben Zimmermann, were found to be functional when tested, meaning any individual who copied them could have potentially gained unauthorized access to sensitive cloud infrastructure and AI services.

The extensive scan, which analyzed 7.25 million public Gitee repositories, revealed that the majority of the exposed keys belonged to cloud providers and AI platforms. Leading the list were 9,473 Alibaba Cloud AccessKey pairs, followed by 3,805 YouTube API keys, 3,170 DeepSeek keys, 1,489 keys for Zhipu's BigModel AI platform, and 1,348 OpenAI keys. Smaller numbers of exposed credentials were also found for major providers like AWS, Google Cloud, and GitHub.

Zimmermann employed two distinct secret-scanning tools, TruffleHog and Kingfisher, to comb through the repositories' full commit histories. Both tools independently verified the validity of the credentials by attempting to use them against their respective services. While TruffleHog identified 12,737 working credentials and Kingfisher found 14,664, only 1,001 were detected by both scanners, highlighting the importance of using multiple tools for comprehensive detection. The scan also revealed that a significant portion of these exposed keys, 12,242 in total, originated from commits made in 2025, with some dating back as far as the late 2000s.

Among the discovered credentials was an "active organization-admin credential for a major enterprise AI platform," according to Zimmermann's report. The specific platform was not named, nor was the extent of the access granted by this credential detailed, underscoring the potential for severe impact. The implications of compromised keys vary by service; for instance, OpenAI notes that a compromised key can lead to data loss, unexpected charges, and service interruptions due to quota exhaustion.

Zimmermann reported his findings to the vendors with the most exposed keys but did not contact the individual developers who leaked them. Crucially, at the time of his report, he had not confirmed that all exposed credentials had been revoked. This means that any developer who pushed a key to a public Gitee repository cannot assume their credentials are safe and should proactively revoke them.

To mitigate the risks, security experts advise treating any key that has ever appeared in a public repository as compromised. This involves revoking the existing key and generating a new one. For Alibaba Cloud users, this means creating a new AccessKey pair, updating applications, and then disabling and deleting the old pair, while also checking for unauthorized activity. Google and YouTube API key holders should rotate their keys and implement API restrictions. OpenAI recommends immediate key rotation and setting spend limits to prevent excessive charges.

This incident serves as a stark reminder of the persistent risks associated with hardcoding sensitive credentials into code, especially in public repositories. While Gitee's public API has limitations that previously hindered security research, this discovery demonstrates that such exposures remain a critical threat vector. The scan's findings also suggest that Gitee's rate of exposed credentials per repository is higher than that observed on GitLab and Bitbucket in previous, similar scans, though direct comparisons are complex due to differing methodologies and tool coverage.

The discovery underscores the need for robust secrets management practices, including the use of environment variables, secure vaults, and automated scanning tools throughout the development lifecycle. Developers must remain vigilant about what information is committed to code repositories, regardless of the platform, to prevent potential breaches and unauthorized access to critical cloud and AI resources.