Gitee, China’s largest indigenous code host, holds 7.25 million public repositories. Ben Zimmermann walked all of them with TruffleHog and MongoDB’s Kingfisher, and every secret he counted had been tested against the service that issued it. Only working credentials made the final list.
The tally is 26,400 verified-live keys. Most unlock cloud and AI services: 9,473 Alibaba Cloud AccessKey pairs, then 3,805 YouTube API keys, 3,170 DeepSeek keys, 1,489 Zhipu BigModel keys and 1,348 OpenAI keys, with smaller clusters for GitHub, AWS, Google Cloud and Gemini.
One entry defies the pattern: an active organisation-admin credential for a major enterprise AI platform. Zimmermann withheld the vendor’s name and did not say what the key could reach.
History keeps what the working tree deletes
Both scanners walked full commit history, so a key stripped from today’s code still counted if it remained live. By commit date, 12,242 came from 2025, nearly half the dataset, while a few trace back to the late 2000s. Roughly one in 766 repositories carried a working Alibaba Cloud key.
Gitee’s API is far thinner than GitHub’s, and it has nothing resembling the platform-wide secret-scanning partnerships GitHub runs with vendors. That gap is why the numbers look the way they do, not a difference in developer habits.
Revoke, do not just delete
Zimmermann sent exposed-key lists to the vendors, but never contacted the developers. He also had no confirmation that any credential had been revoked before publication.
Treat any key that ever touched a public Gitee repository as burned. Delete the old Alibaba AccessKey pair after cutting over, rotate Google and YouTube keys and add API restrictions, and reset OpenAI keys with spend caps in place. Rewriting history fixes nothing.
