Agentic Workflow Hijacking Vulnerability Detection on ask-bonk GitHub Action
7.6M.P.AutoDAN-T
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AutoDAN-TModel=Claude Haiku 4.52026.05 | 7.6 | 6 | |
| AutoDAN-TModel=Gemini-2.5-Flash2026.05 | 4.7 | 5.7 | |
| JAW-NoPromptProvModel=Claude Sonnet 4.52026.05 | 3.3 | 5.7 | |
| JAW-NoCapProfModel=Claude Sonnet 4.52026.05 | 3 | — | |
| Search-basedModel=Claude Sonnet 4.52026.05 | 2.7 | 2.7 | |
| Search-basedModel=Claude Haiku 4.52026.05 | 2.7 | 3.5 | |
| JAWModel=Claude Sonnet 4.52026.05 | 2.3 | 5 | |
| Search-basedModel=Gemini-2.5-Flash2026.05 | 2 | 2 | |
| JAW-NoPromptProvModel=Gemini-2.5-Flash2026.05 | 1.7 | 4 | |
| JAW-NoPromptProvModel=Claude Haiku 4.52026.05 | 1.3 | 3.7 | |
| JAW-NoCapProfModel=Claude Haiku 4.52026.05 | 1.3 | — | |
| JAW-NoCapProfModel=Gemini-2.5-Flash2026.05 | 1.3 | — | |
| JAWModel=Claude Haiku 4.52026.05 | 1 | 3 | |
| JAWModel=Gemini-2.5-Flash2026.05 | 1 | 2.7 |