A scan of 7.25 million public Gitee code folders by researcher Ben Zimmermann uncovered 26,400 active API keys and credentials, including thousands tied to Alibaba Cloud, OpenAI, and DeepSeek.
Independent researcher Ben Zimmermann identified a total of 26.400 active digital keys by scanning 7,25 million public code folders on Gitee, a hosting service widely used in China. The results of the research were published on October 28. Most of the findings related to storage and artificial intelligence services.
Which keys were detected?
The largest category was for Alibaba Cloud, with 9.473 key pairs. This was followed by 3.805 keys for YouTube, 3.170 keys for DeepSeek, 1.489 for Zhipu's BigModel artificial intelligence platform, and 1.348 keys for OpenAI.
According to the researcher, among the findings was an active organization administrator key for a large enterprise artificial intelligence platform, without naming the platform.
How the scan was performed
The research relied on special programs that examined the entire change history of files. This means that even if a developer had later removed a key from the code, it was still visible in the history and counted in the results.
↳
READ ALSO
HACKS.GR / RELATED
Cyber Attacks & Data Breaches
Alpine Linux rebuilt mainframe after hacking attempt
↗
Each key found was tested live on the corresponding service to see if it worked. Of the total findings, 12.242 keys were from the year 2025, while some dated back to the late 2000s.
What happens to exposed keys?
Anyone who finds these digital keys could use them directly. For example, in the case of OpenAI, a lost key allows third parties to use the account's resources without the owner's approval, which can result in unexpected charges.
The researcher forwarded the lists to the respective companies but did not contact the developers who left them exposed. At the time of publication, it was not confirmed whether the exposed keys had been revoked.
G
HACKS.GR / GOOGLE SEARCH
Hacks.gr on Google
Set Hacks.gr as a preferred source.
Preferred source
↗
end of article
Top comments (0)