Jailbreak Attack Defense on PAIR
1ASRSelf-Reminder
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Self-ReminderModel Scale=Qwen3-14B2026.01 | 1 | — | — | — | |
| Self-GuardModel Scale=Qwen3-14B2026.01 | 1 | — | — | — | |
| Self-GuardModel Scale=Qwen3-4B2026.01 | 3 | — | — | — | |
| ReasoningGuardModel Scale=Qwen3-4B2026.01 | 4 | — | — | — | |
| Self-GuardModel Scale=Qwen3-8B2026.01 | 5 | — | — | — | |
| ReasoningGuardModel Scale=Qwen3-8B2026.01 | 6 | — | — | — | |
| Self-ReminderModel Scale=Qwen3-4B2026.01 | 7 | — | — | — | |
| STAR-1Model Scale=Qwen3-8B2026.01 | 7 | — | — | — | |
| STAR-1Model Scale=Qwen3-14B2026.01 | 7 | — | — | — | |
| Self-ReminderModel Scale=Qwen3-8B2026.01 | 9 | — | — | — | |
| ReasoningGuardModel Scale=Qwen3-14B2026.01 | 9 | — | — | — | |
| SafeKeyModel Scale=Qwen3-14B2026.01 | 10 | — | — | — | |
| STAR-1Model Scale=Qwen3-4B2026.01 | 11 | — | — | — | |
| SafeKeyModel Scale=Qwen3-8B2026.01 | 14 | — | — | — | |
| Alpha-steerModel Scale=Qwen3-14B2026.01 | 15 | — | — | — | |
| SafeKeyModel Scale=Qwen3-4B2026.01 | 17 | — | — | — | |
| SafeChainModel Scale=Qwen3-4B2026.01 | 27 | — | — | — | |
| SafeChainModel Scale=Qwen3-14B2026.01 | 27 | — | — | — | |
| InstructModel Scale=Qwen3-14B2026.01 | 37 | — | — | — | |
| SafeChainModel Scale=Qwen3-8B2026.01 | 38 | — | — | — | |
| Alpha-steerModel Scale=Qwen3-8B2026.01 | 40 | — | — | — | |
| Alpha-steerModel Scale=Qwen3-4B2026.01 | 50 | — | — | — | |
| InstructModel Scale=Qwen3-4B2026.01 | 53 | — | — | — | |
| InstructModel Scale=Qwen3-8B2026.01 | 53 | — | — | — | |
| No DefenseModel=R1-Qwen-14B2025.08 | — | — | 68.2 | 97.7 | |
| No DefenseModel=R1-Qwen-32B2025.08 | — | — | 40.9 | 95.5 | |
| ParaphraseModel=R1-Qwen-14B2025.08 | — | — | 59.1 | 100 | |
| ParaphraseModel=R1-Qwen-32B2025.08 | — | — | 54.5 | 97.7 | |
| RealSafe-R1Model=R1-Qwen-14B2025.08 | — | — | 15.9 | 38.6 | |
| RealSafe-R1Model=R1-Qwen-32B2025.08 | — | — | 15.9 | 61.4 | |
| ReasoningGuardModel=R1-Qwen-14B2025.08 | — | — | 15.9 | 50 | |
| ReasoningGuardModel=R1-Qwen-32B2025.08 | — | — | 0 | 22.7 | |
| SafeKeyModel=R1-Qwen-14B2025.08 | — | — | 25 | 86.4 | |
| SAFEPATH-ZSModel=R1-Qwen-14B2025.08 | — | — | 43.2 | 86.4 | |
| SAFEPATH-ZSModel=R1-Qwen-32B2025.08 | — | — | 40.9 | 79.5 | |
| Self-ReminderModel=R1-Qwen-14B2025.08 | — | — | 47.7 | 90.9 | |
| Self-ReminderModel=R1-Qwen-32B2025.08 | — | — | 40.9 | 90.9 | |
| SmoothLLMModel=R1-Qwen-14B2025.08 | — | — | 52.3 | 95.5 | |
| SmoothLLMModel=R1-Qwen-32B2025.08 | — | — | 56.8 | 95.5 | |
| ThinkingIModel=R1-Qwen-14B2025.08 | — | — | 13.6 | 65.9 | |
| ThinkingIModel=R1-Qwen-32B2025.08 | — | — | 18.2 | 90.9 |