Jailbreak Defense on DeepInception
1Harmful ScoreSelf-Reminder
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Self-ReminderModel=Llama22024.02 | 1 | 0.04 | |
| ICDModel=Llama22024.02 | 1 | 0 | |
| SafeDecodingModel=Llama22024.02 | 1 | 0 | |
| Self-ExaminationModel=GPT-4o-mini2025.05 | 1 | 4 | |
| IAModel=GPT-4o-mini2025.05 | 1 | 0 | |
| SAGEModel=GPT-4o-mini2025.05 | 1 | 0 | |
| Self-ReminderModel=GPT-4o2025.05 | 1 | 24 | |
| Goal PrioritizationModel=GPT-4o2025.05 | 1 | 0 | |
| IAModel=GPT-4o2025.05 | 1 | 0 | |
| SAGEModel=GPT-4o2025.05 | 1 | 0 | |
| Self-ReminderModel=Claude-3.5-Sonnet2025.05 | 1 | 0 | |
| ICDModel=Claude-3.5-Sonnet2025.05 | 1 | 2 | |
| Goal PrioritizationModel=Claude-3.5-Sonnet2025.05 | 1 | 0 | |
| IAModel=Claude-3.5-Sonnet2025.05 | 1 | 4 | |
| SAGEModel=Claude-3.5-Sonnet2025.05 | 1 | 0 | |
| SAGEModel=Gemma22025.05 | 1 | 0 | |
| Goal PrioritizationModel=Qwen2.52025.05 | 1 | 2 | |
| IAModel=Qwen2.52025.05 | 1 | 2 | |
| Self-ExaminationModel=Llama3.12025.05 | 1 | 0 | |
| SAGEModel=Llama3.12025.05 | 1 | 0 | |
| Self-ReminderModel=GPT-4o-mini2025.05 | 1.06 | 20 | |
| SafeDecodingModel=Vicuna2024.02 | 1.08 | 0 | |
| Self-ExaminationModel=Llama22024.02 | 1.1 | 0.02 | |
| SAGEModel=Qwen2.52025.05 | 1.1 | 1 | |
| ParaphraseModel=Llama22024.02 | 1.12 | 0.08 | |
| RetokenizationModel=Llama22024.02 | 1.16 | 0.4 | |
| No DefenseModel=Llama22024.02 | 1.18 | 0.1 | |
| PPLModel=Llama22024.02 | 1.18 | 0.1 | |
| Self-ExaminationModel=Qwen2.52025.05 | 1.18 | 6 | |
| Goal PrioritizationModel=GPT-4o-mini2025.05 | 1.19 | 8 | |
| Self-ReminderModel=Qwen2.52025.05 | 1.22 | 4 | |
| No DefenseModel=Claude-3.5-Sonnet2025.05 | 1.24 | 6 | |
| Self-ExaminationModel=Claude-3.5-Sonnet2025.05 | 1.24 | 6 | |
| ICDModel=GPT-4o2025.05 | 1.26 | 8 | |
| Goal PrioritizationModel=Llama3.12025.05 | 1.26 | 12 | |
| ICDModel=GPT-4o-mini2025.05 | 1.4 | 12 | |
| Goal PrioritizationModel=Gemma22025.05 | 1.46 | 42 | |
| Self-ExaminationModel=Gemma22025.05 | 1.88 | 18 | |
| Self-ExaminationModel=GPT-4o2025.05 | 2.18 | 30 | |
| IAModel=Llama3.12025.05 | 2.18 | 6 | |
| IAModel=Gemma22025.05 | 2.22 | 42 | |
| Self-ReminderModel=Llama3.12025.05 | 2.24 | 8 | |
| ICDModel=Qwen2.52025.05 | 2.34 | 60 | |
| Self-ReminderModel=Gemma22025.05 | 2.36 | 24 | |
| ICDModel=Llama3.12025.05 | 2.94 | 28 | |
| No DefenseModel=Llama3.12025.05 | 2.98 | 30 | |
| Self-ExaminationModel=Vicuna2024.02 | 3 | 0.88 | |
| RetokenizationModel=Vicuna2024.02 | 3.16 | 1 | |
| ParaphraseModel=Vicuna2024.02 | 3.6 | 1 | |
| No DefenseModel=Vicuna2024.02 | 3.62 | 1 | |
| PPLModel=Vicuna2024.02 | 3.62 | 1 | |
| Self-ReminderModel=Vicuna2024.02 | 3.66 | 1 | |
| ICDModel=Gemma22025.05 | 3.66 | 74 | |
| No DefenseModel=GPT-4o2025.05 | 3.84 | 72 | |
| ICDModel=Vicuna2024.02 | 3.96 | 1 | |
| No DefenseModel=Qwen2.52025.05 | 4.54 | 92 | |
| No DefenseModel=GPT-4o-mini2025.05 | 4.56 | 94 | |
| No DefenseModel=Gemma22025.05 | 4.76 | 52 |