Adversarial Attack Success on 12 Adversarial Tasks
72.7Attack Success RateEcho Chamber
Evaluation Results
| Method | Links | |
|---|---|---|
| Echo ChamberModel=Gemini 2.5 Flash2026.01 | 72.7 | |
| Echo ChamberModel=Gemini 2.0 Flash2026.01 | 58.3 | |
| Echo ChamberModel=GPT-4.1 mini2026.01 | 54.5 | |
| CrescendoModel=Gemini 2.0 Flash2026.01 | 50 | |
| Echo ChamberModel=GPT-4.12026.01 | 50 | |
| Echo ChamberModel=Gemini 2.5 Pro2026.01 | 45.5 | |
| CrescendoModel=Gemini 2.5 Flash2026.01 | 33.3 | |
| CrescendoModel=Gemini 2.5 Pro2026.01 | 33.3 | |
| DANModel=Gemini 2.5 Flash2026.01 | 25 | |
| DANModel=Gemini 2.0 Flash2026.01 | 25 | |
| CrescendoModel=GPT-4.1 mini2026.01 | 25 | |
| CrescendoModel=GPT-5 nano2026.01 | 25 | |
| Echo ChamberModel=GPT-5 mini2026.01 | 18.2 | |
| CrescendoModel=GPT-4.12026.01 | 16.7 | |
| DANModel=Gemini 2.5 Pro2026.01 | 16.7 | |
| CrescendoModel=GPT-5 mini2026.01 | 16.7 | |
| Echo ChamberModel=GPT-5 nano2026.01 | 16.7 | |
| DANModel=GPT-4.1 mini2026.01 | 0 | |
| DANModel=GPT-4.12026.01 | 0 | |
| DANModel=GPT-5 mini2026.01 | 0 | |
| DANModel=GPT-5 nano2026.01 | 0 |