Containers Don't Keep Secrets: Scanning Docker Hub for Leaked Credentials and Private Keys.
summary
Scan of Docker Hub images processes 200 TB scanned into a BA2 normalized archive, deduplicates layers, and runs format-specific rules to yield ~100,000 candidate secrets. Service validators confirm about 200 validated credentials (81 tied to AI platforms), including a Terraform token granting admin access to 27 organizations. Internet-wide key matching with ZMap/ZGrab links image-embedded private keys to ~200,000 hosts. A local fine-tuned LLM speeds contextual triage and sharply reduces false positives.
AI-written summary · about
chapters (19)
- 0:00Introduction to leaked secrets
- 1:11Supply chain attack risks
- 2:16AI amplifying secret leaks
- 3:01Research gap and approach
- 4:28Selecting Docker namespaces
- 5:55Scanning process overview
- 8:07Raw findings and limitations
- 9:57Validating secrets with API checks
- 10:38Real-world credential examples
- 12:25Build system leaks
- 13:50CI/CD token blind spots
- 15:15Detecting generic secrets
- 17:23AI for secret triage
- 19:51Fine-tuning LLM for secrets
- 24:56Private keys in Docker images
- 26:03Matching keys to internet services
- 28:58Reporting and remediation gaps
- 29:42Key takeaways and conclusion
- 31:11Q&A session
related talks
28mFresh Secrets From the Docks: Lessons Learnt From Analyzing 180,000 Public DockerHub Images
30mBelow the Radar: Identifying Hidden Threats Within the Development Ecosystem
30mWhoops! I Accidentally Leaked My Credentials (again)
33mGaining initial access by exploiting leaked credentials
33mWhoops! I Accidentally Leaked My Cloud Keys
32mContainer Security: Exploits, Vulnerabilities and Defense
35mFollow the data to learn the secret
30mWhat We Learned Dissecting the World’s Most Popular Containers