Public Leaderboard
LeakBench
Weekly security scans of popular open-source AI projects. Ranked by prompt robustness score.
| Project | Score | Status | Scans | Action |
|---|
| Loading leaderboard… |
README Badge
Projects on the leaderboard can add a security score badge to their README. Submit your project below to get listed, then grab your badge from the project page.
Submit a project
Suggest an open-source AI project for the leaderboard.
All scans use Claude Sonnet 4.6 as the target model. Each project's system prompt is extracted from its public repository and tested with 30 adaptive extraction turns and 23 injection probes. Scores are updated weekly.
