Agentic Workflow Hijacking Vulnerability Detection on google-github-actions/run-gemini-cli
6M.P. ScoreJAW-NoPromptProv
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| JAW-NoPromptProvModel=Gemini 2.5 Pro2026.05 | 6 | — | |
| Search-basedModel=Gemini 2.5 Flash2026.05 | 5.5 | — | |
| JAW-NoCapProfModel=Gemini 2.5 Pro2026.05 | 5.3 | 6.3 | |
| JAW-NoCapProfModel=Gemini 2.5 Flash2026.05 | 3 | — | |
| JAW-NoTriggerModel=Gemini 2.5 Flash-Lite2026.05 | 2.3 | 3 | |
| JAW-NoPromptProvModel=Gemini 2.5 Flash2026.05 | 2 | 7 | |
| Search-basedModel=Gemini 2.5 Flash-Lite2026.05 | 2 | — | |
| JAWModel=Gemini 2.5 Pro2026.05 | 1.3 | 6 | |
| AutoDAN-TModel=Gemini 2.5 Flash2026.05 | 1.3 | — | |
| JAWModel=Gemini 2.5 Flash2026.05 | 1.3 | 6 | |
| AutoDAN-TModel=Gemini 2.5 Flash-Lite2026.05 | 1.3 | — | |
| JAW-NoPromptProvModel=Gemini 2.5 Flash-Lite2026.05 | 1.3 | 4 | |
| JAW-NoCapProfModel=Gemini 2.5 Flash-Lite2026.05 | 1.3 | 3.7 | |
| JAWModel=Gemini 2.5 Flash-Lite2026.05 | 1 | 2.3 |