Summary
Independent researcher Ben Zimmermann scanned 7,256,531 public repositories on Gitee, China's largest indigenous Git hosting platform, and found 26,400 credentials that still authenticated at scan time. The results were published on October 28, 2026, including as a guest post on MongoDB's blog developed with the Kingfisher team.
Method
- Two open-source secret scanners, TruffleHog and Kingfisher, read full commit history, so keys that developers later deleted from the latest commit could still be found.
- Each candidate was tested against the vendor's API at scan time, and anything that failed to authenticate was discarded.
- The 26,400 total is deduplicated by raw secret value. TruffleHog found 12,737 unique credentials and Kingfisher found 14,664; both caught 1,001.
What Was Exposed
| Credential type | Count |
|---|---|
| Alibaba Cloud AccessKey pairs | 9,473 |
| YouTube API keys | 3,805 |
| DeepSeek API keys | 3,170 |
| Zhipu BigModel keys | 1,489 |
| OpenAI API keys | 1,348 |
Smaller numbers of GitHub, AWS, Google Cloud and Gemini credentials were also found, along with an active organization-admin credential for a major enterprise AI platform. Nearly half (12,242) came from commits made in 2025; some dated to the late 2000s.
Disclosure
According to The Hacker News, Zimmermann sent vendors with the most exposed keys a list of their affected credentials but did not contact the developers who leaked them. At publication he had not confirmed that any had been revoked, and he did not say whether anyone else had already used the keys.
Why It Matters
Cloud and AI API keys convert directly into spend, data access and model abuse. Because the scanners read git history, a "fixed" repository can still leak a working key. The findings mirror earlier large-scale GitHub secret scans, showing the problem is platform-agnostic.
Recommendations
- Rotate, don't just delete. Any key that has ever been committed should be treated as compromised.
- Scan history of your own repositories with TruffleHog or Kingfisher, including mirrors on alternative platforms.
- Enable push protection and pre-commit secret scanning wherever your developers publish code.
- Prefer short-lived credentials such as workload identity and scoped, expiring tokens over long-lived AccessKeys.
- Set spend and usage alerts on AI and cloud accounts to catch abuse of leaked keys early.